# auray-serverless/model/minimax-m3

`POST https://queue.auray.run/auray-serverless/model/minimax-m3` submits; the response carries a `request_id` and the three URLs to use next.

`POST https://auray.run/auray-serverless/model/minimax-m3` is the same request run synchronously: the connection is held and the response is the model's OUTPUT, the same document the result URL would have answered. A request that usually takes longer than a run may hold its connection is refused `400 model_not_synchronous` there before anything is charged, naming `expected_seconds` and `wait_ceiling_seconds` — submit it to the queue instead. Past the deadline a run is `504 run_timed_out` with `X-Auray-Request-Timeout-Type: user` — the request is still alive, running and charged or, while it waits to start, not yet charged (`charged` says which) — so poll `status_url` rather than resending, or resend with the same `idempotency_key` to be handed this request again instead of a second one. A run holds one of the connections this platform, this product and your account may keep waiting; with all of them in use it is refused `429 wait_capacity` with `Retry-After` before anything is charged (`charged: false`), and retrying after that pause is safe.

One of Auray's catalogue models, billed once in credits. Whether it runs right now is `GET /v1/catalogue`'s `addresses` (no key needed): `connected` runs; `described` answers `501 model_not_connected` and `paused` `503 model_disabled`, both before anything is charged.

Price: Credits per million tokens: 25 for the prompt and 100 for the answer, its reasoning included. A prompt of more than 512000 tokens is charged 50 for the prompt and 200 for the answer, for the whole call. Quoted when you submit at the prompt as estimated — its UTF-8 bytes over 4, 1100 tokens an image — and at the whole answer `settings.maxTokens` asks for, 8192 when it names none. The charge settles to the tokens the model was billed for and never above the quote.

Fields as published for this model.

Every error code is explained at https://auray.ai/docs/api/errors.

Address: auray-serverless/model/minimax-m3
Servers: https://queue.auray.run · https://auray.run
Scope:   text:write
Auth:    Authorization: Bearer auray_sk_…
API enabled: yes

## Input

- inputs.images (array of url, optional) — Array of image inputs (an http(s) URL; upload a local file first and send its read URL). Documented range: 1 or more.
- inputs.videos (array of url, optional) — Array of video inputs (an http(s) URL; upload a local file first and send its read URL). Documented range: 1 or more.
- settings.systemPrompt (string, optional) — System-level instruction that guides the model's behavior and output style across the entire generation. Documented range: 1 to 200,000 characters.
- settings.temperature (number, optional, default: 1) — Controls randomness in generation. Lower values produce more deterministic outputs, higher values increase variation and creativity. Documented range: 0 to 2.
- settings.topP (number, optional, default: 0.95) — Nucleus sampling parameter that controls diversity by limiting the probability mass. Lower values make outputs more focused, higher values increase diversity. Documented range: 0 to 1.
- settings.maxTokens (integer, optional, default: 8192) — Maximum number of tokens to generate in the response. Left out, a call is sent and quoted for the default here — the model's own where it is smaller than 8,192; a larger value, up to the maximum, is quoted for that answer. Documented range: 1 to 512,000.
- messages[].role (string, required, one of: "user", "assistant") — Role of the message author.
- messages[].content (string, required) — Text content of the message. Documented range: 1 or more characters.
- outputFormat (string, optional, default: "TEXT", one of: "TEXT") — Specifies the file format of the generated output. The available values depend on the task type and the specific model's capabilities. - \`JPG\`: Best for photorealistic images with smaller file sizes (no transparency). - \`PNG\`: Lossless compression, supports high quality and transparency (alpha channel). - \`WEBP\`: Modern format providing superior compression and transparency support. - \`MP4\`: Widely supported video container (H.264), recommended for general use. - \`WEBM\`: Optimized for web delivery. - \`MOV\`: QuickTime format, common in professional workflows (Apple ecosystem). - \`GIF\`: Animated image format (no audio), suitable for short loops or previews. - \`MP3\`: Compressed audio, smaller file size. - \`WAV\`: Uncompressed, high-quality audio. - \`FLAC\`: Lossless compression. - \`OGG\`: Open-source compressed audio format (Vorbis codec). - \`SVG\`: Scalable Vector Graphics. - \`TIFF\`: High-quality output supporting layers. - \`GLB\`: Binary glTF format for 3D models, including geometry, textures, and PBR materials. > [!NOTE] > \*\*Transparency\*\*: If you are using features like background removal or LayerDiffuse that require transparency, you must select a format that supports an alpha channel (e.g., \`PNG\`, \`WEBP\`, \`TIFF\`). \`JPG\` does not support transparency.
- includeUsage (boolean, optional) — Include token usage statistics in the response.
- idempotency_key (string, optional) — Your own name for this request. Send the same one twice and the second call is answered with the first request rather than charged again. `Idempotency-Key:` as a header is accepted as an alias — that is the spelling a retrying proxy replays, and the one to use when running a model on `auray.run`, where a request that outlives its connection is answered 504 and recovered by replay. Omit it and the platform mints one.

## Rules

Across fields, as the model's documentation states them; judged before any charge, and warned (Deprecation) until enforcement starts.

- `messages` is required.
- `messages` must hold at least 1 item. In every item of `messages`: `messages[].role` and `messages[].content` are required.

## Response

A submit answers the status document (`status`: IN_QUEUE, IN_PROGRESS or COMPLETED; `request_id`;
`queue_position` while queued; `logs`; `metrics`) and then, beside it:
- product (string) — Which of this platform's products ran it.
- provider_job_id (string) — The id the job runs under, once it has been handed to its model — possibly while still `queued`; null before. Quote it to support.
- model (string) — The address that ran it, echoed back.
- engine (string) — The service's own name for the model that ran, where it reports one — a catalogue model's is its `id` on GET /v1/models; null otherwise. What `model` carried before it carried the address.
- attribution (string) — The credit line the model's licence requires you to show; null where none is required.
- credits_charged (integer) — What this job cost. 0 on a replay, or when a pass or a promotion paid for it (`coverage` says which).
- credits_quoted (integer) — What this job costs, in credits: the price it was quoted at submit. Equal to `credits_charged` once it has been charged; while it waits in the queue to start, `credits_charged` is 0 and this is what it will be charged when it starts (a cancel before then costs nothing) — and a replay of a job still waiting quotes the body that job was queued with, which is the one it runs, not the replay's. For a catalogue model it is the hold, and the settle may give part of it back — a settled read's `credits_charged` says what it finally cost.
- coverage (object) — Present when a pass or a promotion paid for this job — `{"by": "pass", "pass": "seedance-2-5"}` or `{"by": "promotion"}`, and `credits_charged` is 0 — or when you hold a pass for this model that did not cover this job: `{"by": null, "miss": {"pass": "video", "why": "door"}}`, `why` being `door` (the pass does not work here) or `shape` (it does not cover this kind of job). Absent otherwise. Send `X-Auray-Covered-Only: 1` to run a job only if a pass or a promotion pays for it.
- credits_remaining (integer) — The balance after it.
- expected_seconds (integer) — A rough render time, for a progress bar; 0 where nothing is published.
- give_up_after_seconds (integer) — The longest this job may take from its submit: the longest it may wait to start (your X-Auray-Request-Timeout, or the product's queue deadline) plus the longest it may run before it is stopped and refunded. Waiting longer is waiting for nothing; the official clients wait this long by default. For a catalogue model the part it may run is its own reap as its registry row states it (`lifecycle.stall_ceiling_seconds` on GET /v1/models); otherwise these are the published limits: we may lengthen them during an incident, never shorten them.
- replayed (boolean) — `true` when your `idempotency_key` matched a job this platform already had. Nothing was charged and nothing was rendered twice.
- waited (boolean) — Present when `wait` was asked for. The rest of this group comes with it.
- timed_out (boolean) — The wait ran out before the job settled. Still a 2xx, never a 408 or 504 — a library that retries on those would resubmit and you would pay twice. Poll `status_url`.
- wait_declined (string) — No seat was free to hold the connection with, so the request was answered at once. `platform_busy`: no product would have helped. `product_busy`: another product's queue might. `account_busy`: you are already waiting on every seat your plan's lanes allow. The job is running either way; poll it.
- status_url (string) — Poll this.
- response_url (string) — Fetch the result from this once the status is COMPLETED.
- cancel_url (string) — `PUT` to this to ask for a cancellation.
Poll `status_url`, fetch `response_url` once COMPLETED, `PUT` `cancel_url` to stop it.

## Price

Price: Credits per million tokens: 25 for the prompt and 100 for the answer, its reasoning included. A prompt of more than 512000 tokens is charged 50 for the prompt and 200 for the answer, for the whole call. Quoted when you submit at the prompt as estimated — its UTF-8 bytes over 4, 1100 tokens an image — and at the whole answer `settings.maxTokens` asks for, 8192 when it names none. The charge settles to the tokens the model was billed for and never above the quote.

## Paths

POST https://queue.auray.run/auray-serverless/model/minimax-m3
  Run auray-serverless/model/minimax-m3
GET https://queue.auray.run/auray-serverless/model/minimax-m3/requests/{request_id}
  The result
GET https://queue.auray.run/auray-serverless/model/minimax-m3/requests/{request_id}/status
  The status
PUT https://queue.auray.run/auray-serverless/model/minimax-m3/requests/{request_id}/cancel
  Ask for a cancellation

## Example

```bash
curl -X POST https://queue.auray.run/auray-serverless/model/minimax-m3 \
  -H 'Authorization: Bearer auray_sk_…' \
  -H 'Content-Type: application/json' \
  -d '{"messages":[{"role":"user","content":"In one sentence, what does a brass orrery show?"}]}'
```

## Full contract

https://api.auray.ai/v1/models/auray-serverless/model/minimax-m3/openapi.json