AI/ML API

Get Model via GET /model/{id}. Gets one model by id or alias (`gpt-4o` and `openai/gpt-4o` both resolve), with its release date, developer, context length, maximum output, training cutoff and pricing. Answers with or without a credential.

Get Model Metrics via GET /models/metrics. Latency and throughput percentiles (p50/p75/p90/p99 for total duration, time to first token and tokens per second) over a recent window, per model and endpoint. Narrow it with `aliases`. Answers with or without a credential.

Get Batch via GET /v1/batches. Fetches a batch created by Create Batch, by its `id`. An unknown id answers 404 'Inference not found'; omitting the id answers 400 'Query validation error!'.

Get Balance via GET /v1/billing/balance. The account's current credit balance with the time it was last updated, whether it is below the low-balance threshold, and the auto-debit status. A balance of zero is why inference calls answer 403 'You've run out of funds' while management calls keep working.

Get Current API Key via GET /v1/key. The metadata of the key most recently used for inference -- prefix, name, disabled state, model-access scopes, spending limit and monthly usage. No key material is returned. Answers 404 'No used inference API key found' when the account has made no inference call yet, which is an empty result rather than an error.

List API Keys via GET /v1/keys. Lists the account's API keys by PREFIX -- the first 8 characters, the same value the dashboard shows -- with each key's name, whether it is disabled, its model-access scopes, its spending limit and its monthly usage. No key material is returned. This call is itself gated by a per-key endpoint permission: a key without it is refused with 403 'This endpoint is not allowed for this API key', which is a permission to enable at aimlapi.com/app/keys, not a bad key.

List Models via GET /v1/models. Lists every model AI/ML API serves, with its developer, context length, modalities, capabilities and pricing. Filter with `type`, `tags`, `modalities`, `input_modalities`, `output_modalities` or `capabilities`, or name one with `id`. This is how to find a valid `model` value for any inference call -- the catalogue changes often enough that AI/ML API publishes a deprecations feed for it. Answers with or without a credential.

List Model Deprecations via GET /v1/models/deprecations. Lists models that are deprecated or withdrawn, each with the date it was deprecated, the date it shuts down, the reason and the model that replaces it. Check this before pinning a model id: a withdrawn id stops answering. Answers with or without a credential.

Get Model Response via GET /v1/responses/{response_id}. Fetches a response created by Create Model Response, by its id. This is how a `background: true` response is polled. An unknown id answers 404 'Inference not found'.

Get Transcription via GET /v1/stt/{generation_id}. Fetches a transcription started by Create Transcription, by its generation id. An unknown id answers 404 'Inference not found'.

Get Billing Summary via GET /v2/billing. The account's current balance and currency (the v2 billing generation). Shorter than Get Balance; use Get Balance for the low-balance and auto-debit state.

Get Billing Detail via GET /v2/billing/detail. The account holder's id and email beside the current balance, currency and the auto top-up settings (whether it is enabled, the threshold that triggers it and the amount it adds).

List Billing Transactions via GET /v2/billing/transactions. The account's billing ledger: top-ups, charges, bonuses, refunds, adjustments, model usage, expired credits. Filter by `from`/`to` date, `direction` (TOPUP or CHARGE) or `type`; page with `limit` (up to 100) and the `cursor` the previous page returned.

Get Audio Generation via GET /v2/generate/audio. Fetches an audio generation started by Generate Audio, by its `generation_id`. An unknown id answers 404 'Inference not found'.

List Request Logs via GET /v2/logs. The account's per-request log: which model, which key prefix, whether it succeeded, what it cost. Supply either `period` or BOTH `start` and `end` -- with neither, the call is refused with 400 'Query validation error!'. Filter by `model`, `key_prefix` or `status`, and page with `limit` (up to 100) and `offset`.

Get Usage Summary via GET /v2/usage. Total spend and request count over a period. Supply either `period` (for example `24h`) or BOTH `start` and `end` -- with neither, the call is refused with 400 'Query validation error!'. Narrow to one key with `key_prefix`.

Get Usage Detail via GET /v2/usage/detail. Spend and request counts broken down by day, by model, or by both (`group_by`). Supply either `period` or BOTH `start` and `end` -- with neither, the call is refused with 400 'Query validation error!'. `tz` sets the timezone day boundaries are cut on.

Get Video Generation via GET /v2/video/generations. Fetches a video generation started by Generate Video, by its `generation_id`. Video generation is slow, so this is the poll call. An unknown id answers 404 'Inference not found'.

Create Batch via POST /v1/batches. Submits a batch of inference requests for asynchronous processing. `requests` is an ARRAY of items, each with its own `custom_id` and `params`; the call is refused with 400 naming `requests` if it is missing. Returns a batch id to poll with Get Batch.

Cancel Batch via POST /v1/batches/cancel/{batch_id}. Cancels a batch that is still running, by its id. Requests already completed inside the batch are not undone -- this stops the ones that have not run. An unknown id answers 404 'Inference not found'.

Create Chat Completion via POST /v1/chat/completions. The OpenAI-compatible chat completion call, and the main inference entry point: send `model` and `messages`, get a completion. Which further arguments are accepted depends on the model family the `model` id selects -- this schema is the union of every family's, so the provider, not this schema, decides whether `reasoning_effort` or `top_k` means anything for the model you chose, and says so in a 400 naming the exact field. Find a valid `model` with List Models.

Create Embeddings via POST /v1/embeddings. Turns `input` text into embedding vectors with the chosen embedding `model`. `dimensions` and `encoding_format` are accepted by the model families that support them.

Edit Image via POST /v1/images/edits. Edits an existing `image` from a `prompt`, optionally confined to a `mask`. `n`, `size`, `quality`, `background` and `output_format` control the result.

Generate Image via POST /v1/images/generations. Generates images from a `prompt` with the chosen image `model`. The arguments a given model takes vary widely -- `image_size`, `aspect_ratio`, `num_images`, `guidance_scale`, `seed`, `output_format`, safety and acceleration knobs -- and this schema is the union of every image family's; the provider refuses the ones the chosen model does not know, naming the field.

Create Message via POST /v1/messages. The Anthropic-compatible messages call: send `model`, `messages` and `max_tokens`, get a message. `system`, `tools`, `tool_choice`, `thinking`, `stop_sequences` and `metadata` follow Anthropic's own shapes. Use Create Chat Completion for the OpenAI-shaped equivalent.

Run OCR via POST /v1/ocr. Extracts text from a `document` (a URL or base64 payload) with an OCR model. `pages` limits which pages are read and `include_image_base64` returns the page images alongside the text.

Create Model Response via POST /v1/responses. The OpenAI Responses-API call: send `model` and `input`, get a response object. Supports stateful conversation through `previous_response_id`, server-side storage through `store`, hosted tools through `tools`, and `background` for long runs -- poll a background response with Get Model Response.

Create Transcription via POST /v1/stt/create. Starts a transcription of an audio `url` with a speech-to-text `model`. Returns a generation id; fetch the text with Get Transcription. Diarization, punctuation, redaction, summarization and intent/topic detection are available on the model families that support them.

Create Speech via POST /v1/tts. Turns `text` into audio with a text-to-speech `model` and a `voice`. Voice, format, sample-rate and prosody arguments differ by provider family and this schema is their union.

Generate Audio via POST /v2/generate/audio. Starts a music or audio generation from a `prompt` (and `lyrics`, for the models that sing) with an audio `model`. Returns a generation id; fetch the result with Get Audio Generation.

Preprocess Reference Audio via POST /v2/generate/audio/preprocess. Prepares a `reference_audio_url` for the audio models that clone or cover a reference track, returning the handle those models take.

Generate Video via POST /v2/video/generations. Starts a video generation from a `prompt` and/or an `image_url` with a video `model`. Returns a generation id; fetch the result with Get Video Generation. Duration, aspect ratio, resolution, audio and reference-image arguments vary by model family and this schema is their union.