<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
    <id>https://docs.serenityedge.ai/release-notes/</id>
    <title>Serenity Edge Blog</title>
    <updated>2026-11-19T00:00:00.000Z</updated>
    <generator>https://github.com/jpmonette/feed</generator>
    <link rel="alternate" href="https://docs.serenityedge.ai/release-notes/"/>
    <subtitle>Serenity Edge Blog</subtitle>
    <icon>https://docs.serenityedge.ai/img/favicon.ico</icon>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.19]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-19/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-19/"/>
        <updated>2026-11-19T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[Breaking changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="breaking-changes">Breaking changes<a href="https://docs.serenityedge.ai/release-notes/2026-11-19/#breaking-changes" class="hash-link" aria-label="Direct link to Breaking changes" title="Direct link to Breaking changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="two-tools-with-the-same-name-are-refused">Two tools with the same name are refused<a href="https://docs.serenityedge.ai/release-notes/2026-11-19/#two-tools-with-the-same-name-are-refused" class="hash-link" aria-label="Direct link to Two tools with the same name are refused" title="Direct link to Two tools with the same name are refused" translate="no">​</a></h3>
<p>Since 19 November 2026, <code>POST /v1/chat/completions</code> and <code>POST /v1/responses</code> refuse a request that declares two function tools with the same name: HTTP 400 with <code>type</code> <code>invalid_request_error</code> and the message <code>More than one tool is named 'X': each tool needs its own name.</code>. <code>param</code> points at the second tool with that name, <code>tools[&lt;i&gt;].function.name</code> on chat completions and <code>tools[&lt;i&gt;].name</code> on the Responses API, and names are compared exactly, letter case included. Such a request used to be accepted, but the arguments of a call to that name were converted with the schema of the tool declared first, so your application could not tell which tool was meant: give each tool its own name. To fix a refused request, look in your <code>tools</code> for two entries with the same name, and rename or remove one. See <a class="" href="https://docs.serenityedge.ai/docs/tool-calling/#give-each-tool-its-own-name">Give each tool its own name</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="sandboxes">Sandboxes<a href="https://docs.serenityedge.ai/release-notes/2026-11-19/#sandboxes" class="hash-link" aria-label="Direct link to Sandboxes" title="Direct link to Sandboxes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="save-a-sandbox-as-a-template-on-request">Save a sandbox as a template, on request<a href="https://docs.serenityedge.ai/release-notes/2026-11-19/#save-a-sandbox-as-a-template-on-request" class="hash-link" aria-label="Direct link to Save a sandbox as a template, on request" title="Direct link to Save a sandbox as a template, on request" translate="no">​</a></h3>
<p>Set a sandbox up once, save it as a private template of your account with the E2B SDKs' <code>create_snapshot</code>, and create sandboxes from it that start where the saved one was, with its memory and disk. Templates are enabled on request, per account, for accounts that have Sandboxes (see <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/#access">Access</a>). Each GiB a template stores is billed at 0.08 USD per GiB-month, the price of a paused sandbox, and an account keeps up to 10 templates and 20 GiB by default. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/templates/">Save a sandbox as a template</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="known-issue">Known issue<a href="https://docs.serenityedge.ai/release-notes/2026-11-19/#known-issue" class="hash-link" aria-label="Direct link to Known issue" title="Direct link to Known issue" translate="no">​</a></h3>
<p>Deleting a template by the snapshot id that the SDKs return (<code>&lt;your account&gt;/&lt;name&gt;:&lt;tag&gt;</code>) gets HTTP 404 with <code>edge_route_not_available</code>, and <code>delete_snapshot</code> returns <code>False</code>: the template is not deleted. Until the fix, delete by the bare name, without your account's namespace, such as <code>Sandbox.delete_snapshot("data-env")</code>. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/#known-issues-in-the-preview">Known issues in the preview</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-11-19/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">New page: <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/templates/">Save a sandbox as a template</a>, in the Sandboxes section, with its price on <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/pricing-and-usage/#prices">Pricing and usage</a>.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/tool-calling/">Tool calling</a> gains "Give each tool its own name" and "Known limitations", and <a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">Errors and rate limits</a> and the <a class="" href="https://docs.serenityedge.ai/docs/responses-api/#errors">Responses API</a> list the new 400.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.18]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-18/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-18/"/>
        <updated>2026-11-18T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[Legal]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="legal">Legal<a href="https://docs.serenityedge.ai/release-notes/2026-11-18/#legal" class="hash-link" aria-label="Direct link to Legal" title="Direct link to Legal" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="legal-documents-version-13">Legal documents version 1.3<a href="https://docs.serenityedge.ai/release-notes/2026-11-18/#legal-documents-version-13" class="hash-link" aria-label="Direct link to Legal documents version 1.3" title="Direct link to Legal documents version 1.3" translate="no">​</a></h3>
<p>Version 1.3 of the <a class="" href="https://docs.serenityedge.ai/terms/changelog/">legal documents</a> takes effect today and replaces version 1.2. It adds the terms for media models, the models that generate or edit images, generate video or generate speech. The <a class="" href="https://docs.serenityedge.ai/terms/changelog/">Legal document changelog</a> lists what changes, in plain words and document by document:</p>
<ul>
<li class=""><strong>Media models</strong> are covered by Annex 2 of the <a class="" href="https://docs.serenityedge.ai/terms/terms-of-service/">Terms of Service</a> and Exhibit 5 of the <a class="" href="https://docs.serenityedge.ai/terms/data-processing-agreement/">Data Processing Agreement</a>. Nothing changes for the other models.</li>
<li class=""><strong>Generated images and videos are kept in memory for up to 60 minutes</strong> so that you can download them, never on disk, and are then deleted. An image returned in the response itself, and generated speech, are not kept.</li>
<li class=""><strong>Automated safety checks</strong> apply to the prompts, input images and outputs of the image and video models. They keep no copy, and no person reviews them.</li>
<li class=""><strong>Every generated image, video and audio is marked as made by AI</strong>, with marks that do not identify you. Do not remove or alter them on purpose.</li>
<li class=""><strong>Voices and real people.</strong> The speech models' voices are synthetic presets; do not present them as a real person's voice. Do not depict a real, identifiable person, or imitate their voice, without their consent.</li>
<li class=""><strong>Fees</strong> may be stated per second of audio, per character, per image or per second of video, besides tokens.</li>
</ul>
<hr>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.17]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-17/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-17/"/>
        <updated>2026-11-17T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="read-your-sandbox-usage-with-get-v1usage">Read your sandbox usage with <code>GET /v1/usage</code><a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#read-your-sandbox-usage-with-get-v1usage" class="hash-link" aria-label="Direct link to read-your-sandbox-usage-with-get-v1usage" title="Direct link to read-your-sandbox-usage-with-get-v1usage" translate="no">​</a></h3>
<p><code>GET /v1/usage</code> returns the usage records of the <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/">Sandboxes</a> created with the API key you call it with: each interval of each sandbox, with its quantities and its amount, the records your sandbox invoice is computed from, page by page with <code>cursor</code> and <code>limit</code>. Each key reads its own records only, under limits of its own (2 requests in flight and 30 per minute) that never take the budget of your sandbox calls. Inference has no usage records to read. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/pricing-and-usage/#check-your-usage">Check your usage</a> and the <a class="" href="https://docs.serenityedge.ai/docs/api/edge/get-usage/">API reference</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="speech-to-text-on-post-v1audiotranscriptions">Speech to text on <code>POST /v1/audio/transcriptions</code><a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#speech-to-text-on-post-v1audiotranscriptions" class="hash-link" aria-label="Direct link to speech-to-text-on-post-v1audiotranscriptions" title="Direct link to speech-to-text-on-post-v1audiotranscriptions" translate="no">​</a></h3>
<p><code>POST /v1/audio/transcriptions</code> transcribes audio files in the format of OpenAI's speech-to-text API, so the OpenAI SDKs work unchanged: send one file per request as a <code>multipart/form-data</code> form with <code>model=vibevoice-asr</code>. The answer is the transcript as <code>json</code> or <code>text</code>, and <code>verbose_json</code> adds segments with their start, end and speaker. More than 50 languages are detected automatically. Each request takes up to 5 minutes of audio and a body of up to 16 MiB, in FLAC, M4A or MP4, MP3, Ogg, WAV or WebM. See <a class="" href="https://docs.serenityedge.ai/docs/audio-transcriptions/">Speech to text</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="remove-the-tags-for-audio-without-speech">Remove the tags for audio without speech<a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#remove-the-tags-for-audio-without-speech" class="hash-link" aria-label="Direct link to Remove the tags for audio without speech" title="Direct link to Remove the tags for audio without speech" translate="no">​</a></h3>
<p>The speech-to-text model marks audio without speech with tags in square brackets, such as <code>[Silence]</code> or <code>[Noise]</code>, in the text and the segments. They are kept by default. Send <code>remove_nonspeech_tags=true</code> (with the OpenAI SDKs, in <code>extra_body</code>) to remove them from every format; billing does not change. See <a class="" href="https://docs.serenityedge.ai/docs/audio-transcriptions/#non-speech-tags">Non-speech tags</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="tool-calls-cut-by-the-output-limit-end-with-length">Tool calls cut by the output limit end with <code>length</code><a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#tool-calls-cut-by-the-output-limit-end-with-length" class="hash-link" aria-label="Direct link to tool-calls-cut-by-the-output-limit-end-with-length" title="Direct link to tool-calls-cut-by-the-output-limit-end-with-length" translate="no">​</a></h3>
<p>On <code>POST /v1/chat/completions</code>, an answer whose output limit cuts it in the middle of a tool call now ends with <code>finish_reason: "length"</code> instead of <code>"tool_calls"</code>, with the same <code>usage</code>. On <code>POST /v1/responses</code>, such a response ends with <code>status: "incomplete"</code> and <code>incomplete_details.reason: "max_output_tokens"</code>, and a stream ends with a <code>response.incomplete</code> event. Before you run a call, check that <code>finish_reason</code> is <code>"tool_calls"</code> and that <code>usage.completion_tokens</code> is below your <code>max_tokens</code>, and send <code>tools</code> on every request of the conversation. See <a class="" href="https://docs.serenityedge.ai/docs/tool-calling/">Tool calling</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="models">Models<a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#models" class="hash-link" aria-label="Direct link to Models" title="Direct link to Models" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="vibevoice-asr-for-speech-to-text"><code>vibevoice-asr</code> for speech to text<a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#vibevoice-asr-for-speech-to-text" class="hash-link" aria-label="Direct link to vibevoice-asr-for-speech-to-text" title="Direct link to vibevoice-asr-for-speech-to-text" translate="no">​</a></h3>
<p><code>vibevoice-asr</code> (alias <code>microsoft/vibevoice-asr</code>) joins the OpenAI-style list (<code>GET /v1/models?format=openai</code>) with <code>endpoints: ["/v1/audio/transcriptions"]</code> and <code>input_modalities: ["audio"]</code>. It is billed per second of audio, rounded up per request, at $0.000047 per second ($0.1692 per hour), listed in <code>audio_pricing.per_second</code> instead of a price per token. A request that ends in an error is not billed. Current prices are listed live on <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#speech-to-text">Pricing and limits</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-11-17/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/sandboxes/pricing-and-usage/#check-your-usage">Sandbox pricing and usage</a> gains "Check your usage", <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/limits-and-quotas/#request-limits">Limits and quotas</a> the limits of <code>GET /v1/usage</code>, and the API reference a <a class="" href="https://docs.serenityedge.ai/docs/api/edge/get-usage/">Usage</a> section.</li>
<li class="">New guide: <a class="" href="https://docs.serenityedge.ai/docs/audio-transcriptions/">Speech to text</a>, with the model on <a class="" href="https://docs.serenityedge.ai/docs/models/#speech-to-text">Models</a>, its price on <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#speech-to-text">Pricing and limits</a> and the route in the <a class="" href="https://docs.serenityedge.ai/docs/api/edge/create-transcription/">API reference</a>.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.14]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-14/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-14/"/>
        <updated>2026-11-14T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="one-prompt-cache-per-api-key">One prompt cache per API key<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#one-prompt-cache-per-api-key" class="hash-link" aria-label="Direct link to One prompt cache per API key" title="Direct link to One prompt cache per API key" translate="no">​</a></h3>
<p>On <code>POST /v1/chat/completions</code> and <code>POST /v1/responses</code>, cached prompt content is never shared between API keys. A prefix is served from the cache only to requests made with the key that sent it, so the first request of each key with a given prefix is billed at the full input price, even when another key has sent the same prefix. See <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#cached-input">Cached input</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="cache_salt-splits-a-keys-cache"><code>cache_salt</code> splits a key's cache<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#cache_salt-splits-a-keys-cache" class="hash-link" aria-label="Direct link to cache_salt-splits-a-keys-cache" title="Direct link to cache_salt-splits-a-keys-cache" translate="no">​</a></h3>
<p><code>POST /v1/chat/completions</code> takes an optional <code>cache_salt</code>: requests of one key with the same value share one cache, and requests with different values, or without one, do not. A platform that serves many users with one key can give each of them a cache of their own. The value must be a string of 1 to 128 characters without <code>@</code>, <code>/</code>, <code>\</code> or NUL, or the request is refused with HTTP 400 and <code>param: "cache_salt"</code>. The Responses API has no such parameter. See <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#cached-input">Cached input</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="reasoning-effort-levels">Reasoning effort levels<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#reasoning-effort-levels" class="hash-link" aria-label="Direct link to Reasoning effort levels" title="Direct link to Reasoning effort levels" translate="no">​</a></h3>
<p><code>reasoning.effort</code> (and <code>reasoning_effort</code> on chat completions) now works as published. <code>none</code> and <code>minimal</code> turn reasoning off: <code>minimal</code> used to turn it on. On a model with reasoning levels, <code>low</code>, <code>medium</code> and <code>high</code> (or <code>xhigh</code>) are now three levels, from the least reasoning to the most, where all of them used to give the highest; on a model without levels, any of them turns reasoning on. Any other value is ignored, where it used to turn reasoning on. <a class="" href="https://docs.serenityedge.ai/docs/models/#reasoning">Reasoning</a> lists the levels of each model.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="streams-that-end-early-are-billed-for-what-you-received">Streams that end early are billed for what you received<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#streams-that-end-early-are-billed-for-what-you-received" class="hash-link" aria-label="Direct link to Streams that end early are billed for what you received" title="Direct link to Streams that end early are billed for what you received" translate="no">​</a></h3>
<p>A <a class="" href="https://docs.serenityedge.ai/docs/streaming/">stream</a> that ends before its usage chunk, because you closed the connection or because it reached the 600-second limit, is billed for the output you received, on an estimate of its tokens; before, it was not billed. A stream that fails on our side, or that ends before any output, is still not billed. See <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#prices">Prices</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="legal">Legal<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#legal" class="hash-link" aria-label="Direct link to Legal" title="Direct link to Legal" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="legal-documents-version-12">Legal documents version 1.2<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#legal-documents-version-12" class="hash-link" aria-label="Direct link to Legal documents version 1.2" title="Direct link to Legal documents version 1.2" translate="no">​</a></h3>
<p>Version 1.2 of the <a class="" href="https://docs.serenityedge.ai/terms/changelog/">legal documents</a> takes effect today and replaces version 1.1. The <a class="" href="https://docs.serenityedge.ai/terms/changelog/">Legal document changelog</a> lists what changes, in plain words and document by document:</p>
<ul>
<li class=""><strong>Sandboxes are covered</strong> by Annex 1 of the <a class="" href="https://docs.serenityedge.ai/terms/terms-of-service/">Terms of Service</a> and Exhibit 4 of the <a class="" href="https://docs.serenityedge.ai/terms/data-processing-agreement/">Data Processing Agreement</a>: what a sandbox stores and for how long, what is logged about it, and what not to put in a sandbox during the private preview.</li>
<li class=""><strong>Processing locations are stated in general terms.</strong> Inference runs on Substrate AI's own GPUs in facilities in the European Union, and request handling in Microsoft Azure data centres in the European Union. The documents no longer name individual sites or cloud regions.</li>
<li class=""><strong>With a Serenity Star account,</strong> your requests and their answers pass through Serenity Star's API, which authenticates and meters them.</li>
<li class=""><strong>Request metadata</strong> is described by category. It still never includes your prompts or completions, and it is still kept for 90 days.</li>
<li class=""><strong>The websites</strong> <code>serenityedge.ai</code>, with its playground, and <code>docs.serenityedge.ai</code> are both covered by the <a class="" href="https://docs.serenityedge.ai/terms/legal-notice/">Legal notice</a>, the <a class="" href="https://docs.serenityedge.ai/terms/privacy-policy/">Privacy Policy</a> and the <a class="" href="https://docs.serenityedge.ai/terms/cookie-policy/">Cookie Policy</a>.</li>
</ul>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="sandboxes">Sandboxes<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#sandboxes" class="hash-link" aria-label="Direct link to Sandboxes" title="Direct link to Sandboxes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="one-api-host-and-one-key-for-inference-and-sandboxes">One API host and one key for inference and sandboxes<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#one-api-host-and-one-key-for-inference-and-sandboxes" class="hash-link" aria-label="Direct link to One API host and one key for inference and sandboxes" title="Direct link to One API host and one key for inference and sandboxes" translate="no">​</a></h3>
<p>Set <code>E2B_DOMAIN</code> to <code>serenityedge.ai</code> and <code>E2B_API_KEY</code> to your Serenity Edge API key. The E2B SDKs then manage sandboxes through <code>https://api.serenityedge.ai</code>, the host and key you already use for inference, on the E2B API's paths (<code>/sandboxes</code>, <code>/v2/sandboxes</code>), which are not under <code>/v1</code>. See the <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/quickstart/">quickstart</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="sandbox-traffic-goes-to-the-sandbox-domain">Sandbox traffic goes to the sandbox domain<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#sandbox-traffic-goes-to-the-sandbox-domain" class="hash-link" aria-label="Direct link to Sandbox traffic goes to the sandbox domain" title="Direct link to Sandbox traffic goes to the sandbox domain" translate="no">​</a></h3>
<p>A sandbox's commands, files, terminal and ports go to its sandbox domain, which the API now returns with each sandbox in the <code>domain</code> field of the create and connect answers. The SDKs use it for you, and you do not set it. In the preview it is <code>vlc1.serenitysandboxes.cloud</code>. What a sandbox serves is never served under <code>serenityedge.ai</code>. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/#the-sandbox-domain">The sandbox domain</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="request-limits-of-the-control-api">Request limits of the control API<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#request-limits-of-the-control-api" class="hash-link" aria-label="Direct link to Request limits of the control API" title="Direct link to Request limits of the control API" translate="no">​</a></h3>
<p>Each key may have 8 requests in flight to the sandbox control API and send it 120 requests a minute. Over either limit, the answer is <code>429 edge_key_request_concurrency_limit</code> or <code>429 edge_key_request_rate_limit</code>, with <code>Retry-After</code>. These limits are separate from those of the inference API, and from the running sandboxes and creates per minute of your account. A Serenity Edge API key whose account does not have Sandboxes enabled gets <code>403 edge_sandboxes_not_enabled</code>. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/limits-and-quotas/#request-limits">Request limits</a> and <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/errors/">Errors</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="if-you-set-e2b_domain-to-the-sites-domain">If you set <code>E2B_DOMAIN</code> to the site's domain<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#if-you-set-e2b_domain-to-the-sites-domain" class="hash-link" aria-label="Direct link to if-you-set-e2b_domain-to-the-sites-domain" title="Direct link to if-you-set-e2b_domain-to-the-sites-domain" translate="no">​</a></h3>
<p><code>E2B_DOMAIN=vlc1.serenitysandboxes.cloud</code>, the setting given on 12 November, keeps working with the same key. The documented setting is now <code>E2B_DOMAIN=serenityedge.ai</code>: change to it when you next update your configuration. Your code does not change.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-11-14/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/models/#reasoning">Models</a> gains a Reasoning section with the effort values and the levels of each model; <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#cached-input">Pricing and limits</a> describes the cache per key and <code>cache_salt</code>; the <a class="" href="https://docs.serenityedge.ai/docs/api/edge/create-chat-completion/">chat completions reference</a> adds <code>cache_salt</code> and <code>reasoning_effort</code> and the six effort values; the <a class="" href="https://docs.serenityedge.ai/docs/coding-agents/">coding agents guide</a> no longer says that <code>low</code>, <code>medium</code> and <code>high</code> behave the same.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/data-residency/">Data residency and retention</a>: processing locations in general terms, the edge network in the request flow, the Serenity Star route and a Sandboxes section.</li>
<li class="">The Sandboxes pages document one API host for inference and sandboxes, <code>curl</code> examples against <code>https://api.serenityedge.ai</code>, and ports reached on the host the SDKs give (<code>sandbox.get_host(8080)</code>). <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/network-and-security/#hosts-and-domains">Sandbox network access and security</a> gains a section on hosts and domains, <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/errors/">Sandbox errors</a> the codes of the control API, and the Sandboxes pages say where sandboxes run in general terms, in the European Union.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.12]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-12/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-12/"/>
        <updated>2026-11-12T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[Sandboxes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="sandboxes">Sandboxes<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#sandboxes" class="hash-link" aria-label="Direct link to Sandboxes" title="Direct link to Sandboxes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="sandboxes-in-private-preview">Sandboxes in private preview<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#sandboxes-in-private-preview" class="hash-link" aria-label="Direct link to Sandboxes in private preview" title="Direct link to Sandboxes in private preview" translate="no">​</a></h3>
<p>Serenity Edge Sandboxes, isolated microVMs for AI agents, are in private preview, by invitation: Linux machines that you create through an API, run commands in, pause, resume and kill. They are enabled on request for existing Serenity Edge customers and partners; once they are enabled, the Serenity Edge API key you use for inference works for sandboxes too. To ask for access, see <a class="" href="https://docs.serenityedge.ai/docs/access/enterprise/">Enterprise</a> or write to <a href="mailto:contact@substrate.ai" target="_blank" rel="noopener noreferrer" class="">contact@substrate.ai</a>. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/">Sandboxes</a> and the <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/quickstart/">quickstart</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="compatible-with-the-e2b-sdks">Compatible with the E2B SDKs<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#compatible-with-the-e2b-sdks" class="hash-link" aria-label="Direct link to Compatible with the E2B SDKs" title="Direct link to Compatible with the E2B SDKs" translate="no">​</a></h3>
<p>The API is compatible with the E2B API and the official E2B SDKs (<code>pip install e2b</code>, <code>npm install e2b</code>), tested with version 2.63.0. Set <code>E2B_DOMAIN</code> to <code>vlc1.serenitysandboxes.cloud</code> and <code>E2B_API_KEY</code> to your Serenity Edge API key, and <code>Sandbox.create()</code> works with no other change. The preview runs in one site, in the European Union.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="prices">Prices<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#prices" class="hash-link" aria-label="Direct link to Prices" title="Direct link to Prices" translate="no">​</a></h3>
<p>A running sandbox is billed per second, with no minimum, for its reserved resources: 0.0504 USD per vCPU-hour and 0.0162 USD per GiB-hour of memory. CPU time may be shared under load, as on burstable instances. A paused sandbox is billed 0.08 USD per GiB-month actually stored. Billing started on 12 November 2026, and there are no welcome credits. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/pricing-and-usage/">Pricing and usage</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="the-default-template">The default template<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#the-default-template" class="hash-link" aria-label="Direct link to The default template" title="Direct link to The default template" translate="no">​</a></h3>
<p>The preview has one public template, <code>edge-agents</code>, used by default, so <code>Sandbox.create()</code> needs no arguments (over HTTP, its <code>templateID</code> is <code>base</code> or <code>edge-agents</code>). It is Ubuntu 24.04 with tools for agents (Python 3.12 with uv and Jupyter kernels, Node.js 26 LTS, the .NET 10 SDK, compilers, git and the GitHub CLI, LibreOffice and document tools, and a headless Chromium with Playwright), with 2 vCPUs, 2 GiB of memory and about 26 GiB of disk, of which 20 GiB are free. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/api-reference/#templates">Templates</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="limits-network-and-lifetime">Limits, network and lifetime<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#limits-network-and-lifetime" class="hash-link" aria-label="Direct link to Limits, network and lifetime" title="Direct link to Limits, network and lifetime" translate="no">​</a></h3>
<p>Each account may run up to 48 sandboxes at once and create 60 per minute, and a sandbox runs for at most 24 hours; a paused sandbox keeps its files, memory and processes for up to 30 days. Commands, files and an interactive terminal go through the SDKs over HTTPS; there is no SSH access. A sandbox reaches public websites and APIs over HTTP and HTTPS (ports 80 and 443), through an egress proxy that the template sets up, and nothing else; private and internal addresses are refused. A sandbox created with <code>allow_internet_access: false</code> reaches nothing at all, and is billed the same. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/limits-and-quotas/">Limits and quotas</a> and <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/network-and-security/">Network access and security</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="not-in-the-preview">Not in the preview<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#not-in-the-preview" class="hash-link" aria-label="Direct link to Not in the preview" title="Direct link to Not in the preview" translate="no">​</a></h3>
<p>Desktops and computer use are coming soon. Template builds, public sandbox URLs, outbound network rules set by the client, sandbox logs, create fields outside the supported list, credentials added to outbound requests by Serenity Edge, a spending cap and reading usage records through the API are not available in the preview; each is refused with its own <code>edge_</code> error code where a request can ask for it. When the site is full, a create or a resume gets <code>503 sandbox_capacity_unavailable</code>: retry later. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/#preview-status">Preview status</a> and <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/errors/">Errors</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="preview-limitations">Preview limitations<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#preview-limitations" class="hash-link" aria-label="Direct link to Preview limitations" title="Direct link to Preview limitations" translate="no">​</a></h3>
<p>Paused sandboxes are not backed up yet, so they can be lost if the site's storage fails: keep anything you need outside the sandbox. The preview has one site, and its limits and prices may change; changes will be announced here. Until the sandbox domain is on the Public Suffix List, do not rely on cookie isolation between sandbox host names (see <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/network-and-security/#cookies-and-browsers">Cookies and browsers</a>). There is no service level, and the independent security testing of the service is not finished: do not put production secrets or personal data in a preview sandbox.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="known-issues">Known issues<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#known-issues" class="hash-link" aria-label="Direct link to Known issues" title="Direct link to Known issues" translate="no">​</a></h3>
<p>Writing a file with <code>sandbox.files.write</code> to a path under <code>/tmp</code> a second time fails with HTTP 500 ("permission denied"): write files under <code>/home/user</code> until the fix. Sandbox metrics are not available in the preview. See <a class="" href="https://docs.serenityedge.ai/docs/sandboxes/overview/#known-issues-in-the-preview">Known issues in the preview</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-11-12/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">New section "Sandboxes (private preview)": overview, quickstart, API reference, errors, limits and quotas, pricing and usage, and network access and security.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.06]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-06/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-06/"/>
        <updated>2026-11-06T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[Website]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="website">Website<a href="https://docs.serenityedge.ai/release-notes/2026-11-06/#website" class="hash-link" aria-label="Direct link to Website" title="Direct link to Website" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="playground">Playground<a href="https://docs.serenityedge.ai/release-notes/2026-11-06/#playground" class="hash-link" aria-label="Direct link to Playground" title="Direct link to Playground" translate="no">​</a></h3>
<p>The new <a href="https://serenityedge.ai/en/playground/" target="_blank" rel="noopener noreferrer" class="">playground</a> calls the API live from the browser, with no key or sign-up. You can play Blocks and gems against <code>orion-compass</code> on <code>POST /v1/systemone</code>, or pick <code>orion-compass-flash</code> or <code>d3</code> as the rival. It also has four business cases:</p>
<ul>
<li class="">support inbox triage on <code>POST /v1/decisions</code>;</li>
<li class="">expenses against a policy on <code>POST /v1/decisions</code>, where <code>orion-compass-flash</code> reads the image of each receipt;</li>
<li class="">a personal data hunt on <code>POST /v1/redact</code>;</li>
<li class="">exact clause search on <code>POST /v1/embeddings</code> and <code>POST /v1/rerank</code>.</li>
</ul>
<p>Every demo shows the exact request and response, a copy as cURL against the public API, and what its calls cost at published prices.</p>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.11.05]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-11-05/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-11-05/"/>
        <updated>2026-11-05T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="up-to-8-images-per-request">Up to 8 images per request<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#up-to-8-images-per-request" class="hash-link" aria-label="Direct link to Up to 8 images per request" title="Direct link to Up to 8 images per request" translate="no">​</a></h3>
<p>Chat models now read up to 8 images per request, counted over the whole conversation, on <code>POST /v1/chat/completions</code> and <code>POST /v1/responses</code>. The limit depends on the model: 8 on <code>orion-pro</code>, 8 on <code>orion-plus</code>, 8 on <code>qwen3.8-27b</code>, 8 on <code>qwen3.6-35b-a3b</code> and 1 on <code>paddleocr-vl-1.6</code>. <code>GET /v1/models</code> publishes each model's limit as <code>supported_inputs.references.max</code> of its <code>image</code> input modality. See <a class="" href="https://docs.serenityedge.ai/docs/models/#image-input">Image input</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="an-error-code-for-too-many-images">An error code for too many images<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#an-error-code-for-too-many-images" class="hash-link" aria-label="Direct link to An error code for too many images" title="Direct link to An error code for too many images" translate="no">​</a></h3>
<p>A request with more images than the model reads is still refused with HTTP 400 and the same message, but its <code>code</code> is now <code>too_many_images</code> instead of <code>invalid_request_error</code>, on <code>POST /v1/chat/completions</code> and <code>POST /v1/responses</code>. Images are counted over the whole conversation, earlier turns included.</p>
<p>In a long conversation, send only the most recent images, up to the model's limit, and replace the older ones with a short text. See <a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">Errors and rate limits</a>.</p>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="models">Models<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#models" class="hash-link" aria-label="Direct link to Models" title="Direct link to Models" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="no-quantization-or-tokenizer-on-the-orion-ids">No quantization or tokenizer on the Orion ids<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#no-quantization-or-tokenizer-on-the-orion-ids" class="hash-link" aria-label="Direct link to No quantization or tokenizer on the Orion ids" title="Direct link to No quantization or tokenizer on the Orion ids" translate="no">​</a></h3>
<p>In <code>GET /v1/models</code>, the <code>orion-pro</code> and <code>orion-plus</code> entries no longer carry <code>quantization</code> or <code>tokenizer</code>. Both fields are optional in the provider schema. <code>qwen3.8-27b</code>, <code>qwen3.6-35b-a3b</code> and <code>paddleocr-vl-1.6</code> keep them. See <a class="" href="https://docs.serenityedge.ai/docs/api/edge/list-models/">List models</a>.</p>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="opencode">OpenCode<a href="https://docs.serenityedge.ai/release-notes/2026-11-05/#opencode" class="hash-link" aria-label="Direct link to OpenCode" title="Direct link to OpenCode" translate="no">​</a></h3>
<p>The new <a class="" href="https://docs.serenityedge.ai/docs/coding-agents/">Coding agents</a> guide configures <a href="https://opencode.ai/" target="_blank" rel="noopener noreferrer" class="">OpenCode</a> with <code>orion-pro</code> and <code>orion-plus</code> on chat completions. OpenCode does not ask for reasoning by default, so the guide sets <code>reasoningEffort</code> on each model; any value other than <code>none</code> turns reasoning on. Codex CLI keeps its configuration in the <a class="" href="https://docs.serenityedge.ai/docs/responses-api/#codex-cli">Responses API</a> guide.</p>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.10.30]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-10-30/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-10-30/"/>
        <updated>2026-10-30T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="embeddings-usage-counts-each-inputs-own-tokens">Embeddings usage counts each input's own tokens<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#embeddings-usage-counts-each-inputs-own-tokens" class="hash-link" aria-label="Direct link to Embeddings usage counts each input's own tokens" title="Direct link to Embeddings usage counts each input's own tokens" translate="no">​</a></h3>
<p><code>usage.prompt_tokens</code> on <code>POST /v1/embeddings</code> is now the sum of each input's own tokens. Before, the inputs of one request were counted as if each were as long as the longest of them, so a request with inputs of different lengths reported, and was billed for, more tokens than the model read. A batch now costs the same as sending its inputs one by one; a request with a single input is unchanged, and so are the vectors. See <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#embeddings-billing">Usage and billing</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="embeddings-take-text-only">Embeddings take text only<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#embeddings-take-text-only" class="hash-link" aria-label="Direct link to Embeddings take text only" title="Direct link to Embeddings take text only" translate="no">​</a></h3>
<p><code>POST /v1/embeddings</code> takes text only: a string, or an array of strings. Token ids were listed as an input, but they never reached the model, and requests with them got a generic 400. They now get a 400 that says to send the text. If you use LangChain's <code>OpenAIEmbeddings</code>, set <code>check_embedding_ctx_length=False</code>: by default it sends token ids from OpenAI's tokenizer, which this model does not use. See <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#embeddings-errors">Errors</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="serenity-star-account-keys-on-chat-completions">Serenity Star account keys on chat completions<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#serenity-star-account-keys-on-chat-completions" class="hash-link" aria-label="Direct link to Serenity Star account keys on chat completions" title="Direct link to Serenity Star account keys on chat completions" translate="no">​</a></h3>
<p>Keys from a <a class="" href="https://docs.serenityedge.ai/docs/access/serenity-star-account/">Serenity Star account</a> now work on <code>POST /v1/chat/completions</code> and <code>POST /v1/responses</code> with <code>orion-pro</code>, <code>orion-plus</code> and <code>qwen3.6-35b-a3b</code>; before, these requests failed with HTTP 404. The other models answer HTTP 404 with <code>code: "model_not_found"</code> to these keys. Serenity Star bills this usage at the prices of your Serenity Star account.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="codex-cli-on-the-responses-api">Codex CLI on the Responses API<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#codex-cli-on-the-responses-api" class="hash-link" aria-label="Direct link to Codex CLI on the Responses API" title="Direct link to Codex CLI on the Responses API" translate="no">​</a></h3>
<p><a href="https://github.com/openai/codex" target="_blank" rel="noopener noreferrer" class="">Codex CLI</a> now works with <code>POST /v1/responses</code>. The route accepts Codex's <code>client_metadata</code>, which is dropped, and its <code>namespace</code> tools: the function tools of its plugins and MCP servers, grouped. A call to one of them comes back as a <code>function_call</code> with its <code>name</code> and <code>namespace</code>. Hosted tools are still refused, so set <code>web_search = "disabled"</code> in Codex's <code>config.toml</code>. See <a class="" href="https://docs.serenityedge.ai/docs/responses-api/#codex-cli">Codex CLI</a> for the whole configuration.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="conversations-on-the-decisions-api">Conversations on the Decisions API<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#conversations-on-the-decisions-api" class="hash-link" aria-label="Direct link to Conversations on the Decisions API" title="Direct link to Conversations on the Decisions API" translate="no">​</a></h3>
<p><code>POST /v1/decisions</code> takes a whole conversation in <code>input</code>, as OpenAI's Decisions API does: system, developer and assistant messages, function and custom tool calls with their outputs, and <code>input_json</code> parts, besides user messages. The model reads the conversation as a transcript, each item under a label, so you can ask whether an agent followed its instructions or used its tools well. Every item counts as input tokens. An input of user messages only is read as before. See <a class="" href="https://docs.serenityedge.ai/docs/decisions/#conversations">Conversations</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="cached-input-is-reported-and-billed">Cached input is reported and billed<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#cached-input-is-reported-and-billed" class="hash-link" aria-label="Direct link to Cached input is reported and billed" title="Direct link to Cached input is reported and billed" translate="no">​</a></h3>
<p>The chat models now report the part of a prompt served from the prompt cache in <code>usage.prompt_tokens_details.cached_tokens</code> (<code>usage.input_tokens_details.cached_tokens</code> on the Responses API), and that part is billed at the cached input price, as <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#cached-input">Pricing and limits</a> says. Until now no answer reported cached tokens, so prompts were billed entirely at the input price. Every request of a conversation now goes to the server that holds its prefix, so long conversations and agents that resend the same instructions, such as Codex CLI, are mostly billed at the cached price from their second request.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="embeddings-and-rerank-read-up-to-8192-tokens">Embeddings and rerank read up to 8,192 tokens<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#embeddings-and-rerank-read-up-to-8192-tokens" class="hash-link" aria-label="Direct link to Embeddings and rerank read up to 8,192 tokens" title="Direct link to Embeddings and rerank read up to 8,192 tokens" translate="no">​</a></h3>
<p><code>POST /v1/embeddings</code> reads each input up to its first 8,192 tokens instead of 512, and <code>POST /v1/rerank</code> reads each document with the query up to 8,192 tokens; longer texts are still cut without an error, and what is cut is not billed. A request now takes at most about 300,000 tokens, estimated as a quarter of its characters (on rerank, the query counted once per document); a larger one gets HTTP 400 with <code>param</code> <code>input</code> or <code>documents</code>. See <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#embeddings-limits">Limits</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-10-30/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#embeddings-limits">Embeddings, rerank and redaction</a> guide, <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#embeddings-rerank-and-redaction">Pricing and limits</a> and the API reference say that each embeddings input is read up to its first 8,192 tokens (the rest is ignored and not billed) and that a request takes at most about 300,000 tokens, and no longer list token ids as an input.</li>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/access/serenity-star-account/">Serenity Star account</a> page lists the models these keys can use and says that Serenity Star bills them at your account's prices; <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/">Pricing and limits</a> says its prices are those of Serenity Edge API keys.</li>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/responses-api/#codex-cli">Responses API</a> guide has a Codex CLI section.</li>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/decisions/#conversations">Decisions</a> guide has a Conversations section, and the API reference lists the conversation items <code>input</code> takes.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#cached-input">Pricing and limits</a> says that the prompt cache works in blocks, from which length a prefix is cached, and that a conversation stays on the server that holds its prefix.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.10.29]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-10-29/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-10-29/"/>
        <updated>2026-10-29T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-10-29/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="images-on-d3">Images on d3<a href="https://docs.serenityedge.ai/release-notes/2026-10-29/#images-on-d3" class="hash-link" aria-label="Direct link to Images on d3" title="Direct link to Images on d3" translate="no">​</a></h3>
<p><code>d3</code> now reads images together with the text, like <code>orion-compass-flash</code>: <code>input_image</code> parts on <code>POST /v1/decisions</code>, and <code>images</code> on the Jev-compatible <code>POST /v1/systemone</code>. The limits are the same: base64 data URLs of PNG, JPEG or WebP, at most 4 per request, 8 MiB in total once decoded, 16 megapixels each, and a request body of up to 12 MiB. See <a class="" href="https://docs.serenityedge.ai/docs/decisions/#images">Images</a>.</p>
<p>The tokens the images expand to are input tokens, billed at the model's input price, and <code>usage</code> splits <code>input_tokens</code> into <code>text_tokens</code> and <code>image_tokens</code>. <code>orion-compass</code> still reads text only.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-10-29/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/decisions/">Decisions</a> guide and the API reference list <code>d3</code> among the models that read images.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.10.28]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-10-28/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-10-28/"/>
        <updated>2026-10-28T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-10-28/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="images-on-orion-compass-flash">Images on Orion Compass Flash<a href="https://docs.serenityedge.ai/release-notes/2026-10-28/#images-on-orion-compass-flash" class="hash-link" aria-label="Direct link to Images on Orion Compass Flash" title="Direct link to Images on Orion Compass Flash" translate="no">​</a></h3>
<p><code>orion-compass-flash</code> now reads images together with the text. On <code>POST /v1/decisions</code>, send them as <code>input_image</code> parts. On the Jev-compatible <code>POST /v1/systemone</code>, send them as <code>images</code>.</p>
<p>Images go as base64 data URLs: PNG, JPEG or WebP, at most 4 per request, 8 MiB in total once decoded, and 16 megapixels each. The request body may be up to 12 MiB on this model. See <a class="" href="https://docs.serenityedge.ai/docs/decisions/#images">Images</a>.</p>
<p>The tokens the images expand to are input tokens, billed at the model's input price: <code>usage.input_tokens</code> counts them, and <code>usage.input_tokens_details.image_tokens</code> (<code>usage.image_tokens</code> on <code>POST /v1/systemone</code>) shows their part, next to <code>text_tokens</code>. <code>orion-compass</code> and <code>d3</code> still read text only.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-10-28/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/decisions/">Decisions</a> guide and the API reference describe image input, its limits and how image tokens are billed.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.10.27]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-10-27/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-10-27/"/>
        <updated>2026-10-27T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-10-27/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="responses-api">Responses API<a href="https://docs.serenityedge.ai/release-notes/2026-10-27/#responses-api" class="hash-link" aria-label="Direct link to Responses API" title="Direct link to Responses API" translate="no">​</a></h3>
<p><code>POST /v1/responses</code> serves OpenAI's Responses API on every chat model, with the same keys, prices and limits as chat completions. It covers:</p>
<ul>
<li class="">text and image input;</li>
<li class="">function tools;</li>
<li class="">structured outputs (<code>text.format</code>);</li>
<li class="">reasoning;</li>
<li class="">streaming with Responses events.</li>
</ul>
<p>The OpenAI SDKs' <code>client.responses.create</code>, <code>client.responses.stream</code> and <code>client.responses.parse</code> work as they are.</p>
<p>The API is stateless, since Serenity Edge stores no prompts or answers. Send the whole conversation in <code>input</code> on every request. <code>store: true</code>, <code>previous_response_id</code>, <code>conversation</code> and <code>background</code> are refused with HTTP 400, and responses cannot be retrieved. See the <a class="" href="https://docs.serenityedge.ai/docs/responses-api/">Responses API guide</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="decision-request-limits">Decision request limits<a href="https://docs.serenityedge.ai/release-notes/2026-10-27/#decision-request-limits" class="hash-link" aria-label="Direct link to Decision request limits" title="Direct link to Decision request limits" translate="no">​</a></h3>
<p><code>POST /v1/decisions</code> and the Jev-compatible <code>POST /v1/systemone</code> accept at most 256 questions per request; <code>POST /v1/decisions</code> also takes at most 255 options per <code>choice</code> question and 10 levels per <code>score</code> question. A larger request is refused, with HTTP 400 (422 on <code>POST /v1/systemone</code>), a message that says which limit was hit, and the parameter at fault in <code>param</code>. See <a class="" href="https://docs.serenityedge.ai/docs/decisions/#current-limitations">Current limitations</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="clearer-validation-errors-on-decisions">Clearer validation errors on decisions<a href="https://docs.serenityedge.ai/release-notes/2026-10-27/#clearer-validation-errors-on-decisions" class="hash-link" aria-label="Direct link to Clearer validation errors on decisions" title="Direct link to Clearer validation errors on decisions" translate="no">​</a></h3>
<p>On <code>orion-compass</code>, <code>orion-compass-flash</code> and <code>d3</code>, the error for a malformed decision request now names the field and how to fix it, and the three models refuse the same requests on <code>POST /v1/systemone</code> with the same messages. For example, a choice given as a string gets "each choice must be an object with a 'value' (string or boolean)", and a question whose text is in <code>question</code> instead of <code>instructions</code> is told so. <code>param</code> points at the exact element, such as <code>questions[0].choices[0]</code>. See <a class="" href="https://docs.serenityedge.ai/docs/decisions/#errors">Errors</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-10-27/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">New guide: <a class="" href="https://docs.serenityedge.ai/docs/responses-api/">Responses API</a>.</li>
<li class="">The <a class="" href="https://docs.serenityedge.ai/docs/api/edge/create-response/">API reference</a> documents <code>POST /v1/responses</code>.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.10.14]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-10-14/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-10-14/"/>
        <updated>2026-10-14T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="decision-model-d3">Decision model d3<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#decision-model-d3" class="hash-link" aria-label="Direct link to Decision model d3" title="Direct link to Decision model d3" translate="no">​</a></h3>
<p><code>d3</code> (alias <code>vllm-sr/d3</code>) joins the <a class="" href="https://docs.serenityedge.ai/docs/decisions/#models">decision models</a> in public preview. It is vLLM Semantic Router's open-weight 27B decision model (Apache-2.0) and first on the text board of the Jev Decision Index. It costs $0.042 per million input tokens, and output is not billed. It reads text only, has a context of 24,576 tokens, and is served on both <code>POST /v1/decisions</code> and <code>POST /v1/systemone</code>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="embeddings">Embeddings<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#embeddings" class="hash-link" aria-label="Direct link to Embeddings" title="Direct link to Embeddings" translate="no">​</a></h3>
<p>New <code>POST /v1/embeddings</code> takes OpenAI's embeddings request, so <code>client.embeddings.create</code> in the OpenAI SDKs works as it is: vectors of 1024 dimensions for up to 512 inputs per request, billed on <code>usage.prompt_tokens</code>. Use a Serenity Edge API key. See <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#embeddings">Embeddings, rerank and redaction</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="rerank">Rerank<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#rerank" class="hash-link" aria-label="Direct link to Rerank" title="Direct link to Rerank" translate="no">​</a></h3>
<p>New <code>POST /v1/rerank</code> scores up to 1000 documents against a query, in the format of the rerank APIs of Cohere and Jina. Its tokens are estimated as <code>ceil((characters(query) + characters(document)) / 4)</code> per document, with characters counted as Unicode code points, and returned in <code>usage</code>. See <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#rerank">Rerank</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="redaction">Redaction<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#redaction" class="hash-link" aria-label="Direct link to Redaction" title="Direct link to Redaction" translate="no">​</a></h3>
<p>New <code>POST /v1/redact</code> finds and masks personal data in up to 256 texts (192 KiB per request) in the privacy filter's own format, with <code>model</code> optional; the text sent is processed in memory only. Its tokens are estimated as <code>ceil(characters / 4)</code> over all the inputs. See <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/#redaction">Redaction</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="access-and-errors-on-the-new-routes">Access and errors on the new routes<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#access-and-errors-on-the-new-routes" class="hash-link" aria-label="Direct link to Access and errors on the new routes" title="Direct link to Access and errors on the new routes" translate="no">​</a></h3>
<p>The three routes take Serenity Edge API keys; any other key gets HTTP 403 <code>permission_denied</code> with a message that names the route's API. A model sent to a route that does not serve it, or a request the model cannot process (<code>The model server rejected the request. Check the input sizes and parameters.</code>), gets HTTP 400. See <a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">Errors and rate limits</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="models">Models<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#models" class="hash-link" aria-label="Direct link to Models" title="Direct link to Models" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="embeddings-rerank-and-redaction-models">Embeddings, rerank and redaction models<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#embeddings-rerank-and-redaction-models" class="hash-link" aria-label="Direct link to Embeddings, rerank and redaction models" title="Direct link to Embeddings, rerank and redaction models" translate="no">​</a></h3>
<p>Three models join the OpenAI-style list (<code>GET /v1/models?format=openai</code>), each with its route in <code>endpoints</code> and billed on input tokens only:</p>
<table><thead><tr><th>Model</th><th>Alias</th><th>Route</th><th style="text-align:right">Price per million input tokens</th></tr></thead><tbody><tr><td><code>qwen3-embedding-0.6b</code></td><td><code>qwen/qwen3-embedding-0.6b</code></td><td><code>POST /v1/embeddings</code></td><td style="text-align:right">$0.008</td></tr><tr><td><code>qwen3-reranker-0.6b</code></td><td><code>qwen/qwen3-reranker-0.6b</code></td><td><code>POST /v1/rerank</code></td><td style="text-align:right">$0.008</td></tr><tr><td><code>privacy-filter</code></td><td><code>openai/privacy-filter</code></td><td><code>POST /v1/redact</code></td><td style="text-align:right">$0.009</td></tr></tbody></table>
<p>Current prices are listed live on <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#embeddings-rerank-and-redaction">Pricing and limits</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="paddleocr-vl-16-for-document-pages"><code>paddleocr-vl-1.6</code> for document pages<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#paddleocr-vl-16-for-document-pages" class="hash-link" aria-label="Direct link to paddleocr-vl-16-for-document-pages" title="Direct link to paddleocr-vl-16-for-document-pages" translate="no">​</a></h3>
<p><code>paddleocr-vl-1.6</code> (aliases <code>baidu/paddleocr-vl-1.6</code> and <code>paddlepaddle/paddleocr-vl-1.6</code>) joins the chat models: send one page image with a task prompt (<code>OCR:</code>, <code>Table Recognition:</code>, <code>Formula Recognition:</code> or <code>Chart Recognition:</code>) and it returns the page's text, table, formula or chart data, at $0.016 per million prompt tokens and $0.016 per million completion tokens (a full page is about 1,250 prompt tokens). It reads one image per request, uses temperature 0 unless you send one, has a context of 16,384 tokens and at most 8,192 output tokens, and does not reason, call tools or produce structured outputs. It is available on the API now, and ready for the <a class="" href="https://docs.serenityedge.ai/docs/access/openrouter/">OpenRouter onboarding</a>, which is not live yet; see <a class="" href="https://docs.serenityedge.ai/docs/models/#document-ocr">Document OCR</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-10-14/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">New guide: <a class="" href="https://docs.serenityedge.ai/docs/embeddings-rerank-and-redaction/">Embeddings, rerank and redaction</a>.</li>
<li class="">New page: <a class="" href="https://docs.serenityedge.ai/docs/access/channels/">Routers and marketplaces</a>. Serenity Edge is available as a provider on cortecs and EUrouter, with that platform's own account, key and billing; OpenRouter and other platforms are in onboarding, and the <a class="" href="https://docs.serenityedge.ai/docs/access/openrouter/">OpenRouter page</a> now describes that status.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/models/#document-ocr">Models</a>: document OCR with <code>paddleocr-vl-1.6</code>, and a live table of the embeddings, rerank and redaction models.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/#embeddings-rerank-and-redaction">Pricing and limits</a>: prices, token estimates and limits of the three routes.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">Errors and rate limits</a>: the 400 for a model on the wrong route or rejected by the model server, the 403 that names the route's API, and the 192 KiB limit of <code>POST /v1/redact</code>.</li>
<li class="">API reference: <a class="" href="https://docs.serenityedge.ai/docs/api/edge/create-embedding/">Create embeddings</a>, <a class="" href="https://docs.serenityedge.ai/docs/api/edge/rerank-documents/">Rerank documents</a> and <a class="" href="https://docs.serenityedge.ai/docs/api/edge/redact-text/">Redact personal data</a>, and the <code>endpoints</code> of the new models in <a class="" href="https://docs.serenityedge.ai/docs/api/edge/list-models/">List models</a>.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.10.13]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-10-13/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-10-13/"/>
        <updated>2026-10-13T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[API changes]]></summary>
        <content type="html"><![CDATA[<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="api-changes">API changes<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#api-changes" class="hash-link" aria-label="Direct link to API changes" title="Direct link to API changes" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="decisions-in-preview">Decisions, in preview<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#decisions-in-preview" class="hash-link" aria-label="Direct link to Decisions, in preview" title="Direct link to Decisions, in preview" translate="no">​</a></h3>
<p>New <code>POST /v1/decisions</code> (the format of OpenAI's Decisions API) and <code>POST /v1/systemone</code> (compatible with Jev's format) return typed, calibrated answers (probabilities, choices and scores) instead of generated text. Use a Serenity Edge API key. See <a class="" href="https://docs.serenityedge.ai/docs/decisions/">Decisions</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="image-input-on-the-chat-models">Image input on the chat models<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#image-input-on-the-chat-models" class="hash-link" aria-label="Direct link to Image input on the chat models" title="Direct link to Image input on the chat models" translate="no">​</a></h3>
<p><code>POST /v1/chat/completions</code> accepts up to 4 images per request as base64 data URLs in <code>image_url</code> parts, billed as input tokens. Image URLs are refused with HTTP 400. See <a class="" href="https://docs.serenityedge.ai/docs/models/#image-input">Models</a>.</p>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="server-errors-are-http-500">Server errors are HTTP 500<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#server-errors-are-http-500" class="hash-link" aria-label="Direct link to Server errors are HTTP 500" title="Direct link to Server errors are HTTP 500" translate="no">​</a></h3>
<p>Errors with code <code>upstream_error</code> and <code>timeout</code> now answer HTTP 500 instead of 502 and 504, with the same JSON body and code, so the error object always reaches you. Handle them as before: retry with backoff. See <a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">Errors and rate limits</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="models">Models<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#models" class="hash-link" aria-label="Direct link to Models" title="Direct link to Models" translate="no">​</a></h2>
<h3 class="anchor anchorTargetStickyNavbar_Vzrq" id="orion-compass-decision-models">Orion Compass decision models<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#orion-compass-decision-models" class="hash-link" aria-label="Direct link to Orion Compass decision models" title="Direct link to Orion Compass decision models" translate="no">​</a></h3>
<p><code>orion-compass</code> and <code>orion-compass-flash</code> join the catalogue in preview, billed per input token only; output is free. Prices are listed live on <a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/">Pricing and limits</a>.</p>
<hr>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-10-13/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class="">New guide: <a class="" href="https://docs.serenityedge.ai/docs/decisions/">Decisions</a>.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/models/#image-input">Models</a>: image input.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">Errors and rate limits</a>: 422 on <code>/v1/systemone</code>, 403 for keys that cannot use decisions, and 500 for <code>upstream_error</code> and <code>timeout</code>.</li>
<li class="">API reference: the two decision endpoints, image parts in user messages, and the error responses.</li>
</ul>]]></content>
    </entry>
    <entry>
        <title type="html"><![CDATA[Release 2026.09.22]]></title>
        <id>https://docs.serenityedge.ai/release-notes/2026-09-22/</id>
        <link href="https://docs.serenityedge.ai/release-notes/2026-09-22/"/>
        <updated>2026-09-22T00:00:00.000Z</updated>
        <summary type="html"><![CDATA[First publication of the Serenity Edge documentation.]]></summary>
        <content type="html"><![CDATA[<p>First publication of the Serenity Edge documentation.</p>
<h2 class="anchor anchorTargetStickyNavbar_Vzrq" id="documentation">Documentation<a href="https://docs.serenityedge.ai/release-notes/2026-09-22/#documentation" class="hash-link" aria-label="Direct link to Documentation" title="Direct link to Documentation" translate="no">​</a></h2>
<ul>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/getting-started/">Getting started</a> with curl, Python, Node.js and .NET examples using the OpenAI SDKs against <code>https://api.serenityedge.ai/v1</code>.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/models/">Models</a>: the open Qwen models and the Orion family, with context and output limits.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/pricing-and-limits/">Pricing and limits</a>, including cached input and rate-limit behaviour.</li>
<li class="">Guides on <a class="" href="https://docs.serenityedge.ai/docs/streaming/">streaming</a>, <a class="" href="https://docs.serenityedge.ai/docs/tool-calling/">tool calling</a>, <a class="" href="https://docs.serenityedge.ai/docs/structured-outputs/">structured outputs</a> and <a class="" href="https://docs.serenityedge.ai/docs/errors-and-rate-limits/">errors and rate limits</a>.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/data-residency/">Data residency and retention</a>.</li>
<li class="">Access routes: <a class="" href="https://docs.serenityedge.ai/docs/access/openrouter/">OpenRouter</a>, <a class="" href="https://docs.serenityedge.ai/docs/access/serenity-star-account/">Serenity Star account</a> and <a class="" href="https://docs.serenityedge.ai/docs/access/enterprise/">Enterprise</a>.</li>
<li class=""><a class="" href="https://docs.serenityedge.ai/docs/api/edge/serenity-edge-api/">API reference</a> generated from the OpenAPI specification.</li>
<li class="">Legal documents (version 1.0, effective 22 September 2026), starting with the <a class="" href="https://docs.serenityedge.ai/terms/terms-of-service/">Terms of Service</a>.</li>
</ul>]]></content>
    </entry>
</feed>