API Reference

Authentication

All API requests require a valid API key passed in the Authorization header.

API keys start with ak- and are 48 characters long. Create keys in your dashboard.

Chat Completions

Create a chat completion with the specified model. Supports streaming via Server-Sent Events.

Parameters

Request

POST https://api.apexapi.dev/v1/chat/completions
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{
  "model": "openai/gpt-4o",
  "messages": [
    { "role": "system", "content": "You are a helpful assistant." },
    { "role": "user", "content": "What is the capital of France?" }
  ],
  "temperature": 0.7,
  "max_tokens": 1024,
  "stream": false
}

Response

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1710000000,
  "model": "openai/gpt-4o",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "The capital of France is Paris."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 25,
    "completion_tokens": 8,
    "total_tokens": 33
  }
}

Where OpenAI compatibility stops

Point an OpenAI client at our base URL and it works. These are the four places our behaviour differs, so you find them here instead of in production.

Every stream ends with an extra usage frame

Before data: [DONE] we send one chunk with an empty choices array, an OpenAI-shaped usage object, and an x_apexapi object carrying the request id, the cost in USD and your remaining balance. It is the only way a streaming caller learns what a request cost. The official SDKs ignore a chunk with no choices; a hand-rolled parser that reads choices[0] without checking the length needs a guard.

An empty balance returns 402

A status OpenAI never sends. Every billed route uses it, with error.type set to insufficient_credits_error.

Some parameters do not reach every provider

Models served through Anthropic are not OpenAI-shaped upstream, so a few fields have no equivalent to translate into.

Where a parameter would change the shape of what you get back, we return 400 with the field named in error.param rather than answer you something you cannot parse. Where it only affects sampling, the request runs and the response carries an x-apexapi-unsupported-params header listing what was not applied. Nothing is dropped in silence.

Images & PDFs in Chat

Chat completions accept multimodal content. Instead of a string, a message's content can be an array of parts. It's the same format as the OpenAI API, so existing SDKs work unchanged. The playground supports both via the attach button.

Content part types

Model support

Images work on all vision-capable models. PDFs are supported on Anthropic Claude, OpenAI GPT-4o / GPT-4.1 / o-series, and Google Gemini models. Sending a PDF to a model without PDF support returns 400 with error code unsupported_content.

Attachments are billed as input tokens by the upstream provider, roughly 1,000–1,600 tokens per image and 1,500–3,000 tokens per PDF page. They are processed in transit only and never stored by ApexApi.

Vision request

POST https://api.apexapi.dev/v1/chat/completions
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{
  "model": "anthropic/claude-sonnet-4.6",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is in this image?" },
        {
          "type": "image_url",
          "image_url": { "url": "data:image/jpeg;base64,/9j/4AAQ..." }
        }
      ]
    }
  ]
}

PDF request

POST https://api.apexapi.dev/v1/chat/completions
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{
  "model": "openai/gpt-4o",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "Summarize this document." },
        {
          "type": "file",
          "file": {
            "filename": "report.pdf",
            "file_data": "data:application/pdf;base64,JVBERi0xLjQ..."
          }
        }
      ]
    }
  ]
}

Image Generation

Generate images from text prompts using supported image models.

Parameters

Request

POST https://api.apexapi.dev/v1/images/generations
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{
  "model": "openai/gpt-image-2",
  "prompt": "A serene mountain landscape at sunset",
  "n": 1,
  "size": "1024x1024",
  "quality": "standard"
}

Response

{
  "created": 1710000000,
  "data": [
    {
      "url": "https://uploads.apexapi.dev/generated/req-id-0.png"
    }
  ]
}

The url is always a hosted HTTPS link (never inline base64), whichever model serves the request. Links are temporary. Download the image promptly rather than hotlinking.

Async: submit and poll

The route above holds the connection open for the whole render, which can take a minute or two. If you call it from somewhere with a shorter request timeout, a serverless function or a Lambda on its 30 second default or most HTTP clients out of the box, the connection drops mid-render. The render does not stop: the image finishes and is billed, and you only see a timeout. Use the async rail instead. Same body, same models, same price, but you get a job id immediately and fetch the result on your own schedule.

# Submit, get a job id back right away:
POST https://api.apexapi.dev/v1/images/jobs
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{ "model": "openai/gpt-image-2", "prompt": "A serene mountain landscape at sunset" }

# 202 Accepted:
{ "id": "b71c2f40-...", "status": "queued", "predicted_usd": 0.04 }

# Poll it whenever, from anywhere:
GET https://api.apexapi.dev/v1/images/jobs/b71c2f40-...
Authorization: Bearer ak-your-api-key

{ "id": "b71c2f40-...", "status": "completed",
  "urls": ["https://uploads.apexapi.dev/generated/b71c2f40-0.png"],
  "predicted_usd": 0.04, "actual_usd": 0.04 }

Statuses are queued, processing, completed and failed. As with video, actual_usd stays null until the job settles and stays null on a failed job, because a failed job is not charged.

Video Generation

Generate video from a text prompt (and optionally a starting image) using supported video models like Veo, Kling, Seedance, Hailuo, and Grok Imagine. Video generation is asynchronous: you submit a job and get an id, then poll until it's done (renders take minutes).

Parameters

Submit

POST https://api.apexapi.dev/v1/videos/generations
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{
  "model": "google/veo3.1/lite",
  "prompt": "A calm ocean wave rolling onto a sandy beach at sunset",
  "duration": 4,
  "resolution": "720p",
  "aspect_ratio": "16:9"
}

{
  "id": "a5933807-c44f-4eec-b25a-15f27a19b9ea",
  "status": "queued"
}

Poll for the result

GET https://api.apexapi.dev/v1/videos/generations/a5933807-...
Authorization: Bearer ak-your-api-key

# while rendering (poll every few seconds, renders take minutes):
{ "id": "a5933807-...", "status": "processing",
  "predicted_usd": 0.3, "actual_usd": null }

# when finished:
{ "id": "a5933807-...", "status": "completed", "url": "https://.../video.mp4",
  "predicted_usd": 0.3, "actual_usd": 0.3 }

# on failure (the pre-auth hold is released, you are not charged):
{ "id": "a5933807-...", "status": "failed", "error": "...",
  "predicted_usd": 0.3, "actual_usd": null }

You're billed per second of generated video, matching the duration you requested. A job that fails releases its hold. You are not charged for a video you didn't get.

Every poll response carries the cost in real dollars. predicted_usd is what was held up front, actual_usd is what you were charged once the job settled. It stays null until then, and on a failed job it stays null because nothing was charged.

File Uploads

Host a local image or video so it can be used as a generation input (image_urls / video_url). Two steps: presign, then PUT the raw bytes straight to storage. Uploading is free and never touches your credit balance, but the account must have a positive balance. Files are deleted automatically after about 48 hours, so upload as part of the generation flow, not for long-term storage.

Parameters

POST https://api.apexapi.dev/v1/uploads
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{ "content_type": "video/mp4", "content_length": 8388608 }

{
  "upload_url": "https://...r2.cloudflarestorage.com/...signed...",
  "public_url": "https://uploads.apexapi.dev/refs/1a2b3c4d5e6f7a8b/9f...e2.mp4",
  "key": "refs/1a2b3c4d5e6f7a8b/9f...e2.mp4",
  "expires_in": 600,
  "note": "Files are retained for 48 hours. Use the public_url promptly as a reference input (image_urls / video_url)."
}

PUT {upload_url}
Content-Type: video/mp4

<raw file bytes>

Then pass public_url as image_urls / video_url in a generation request. The daily upload allowance scales with your credit balance (500MB at the base, up to effectively unlimited for high-balance accounts); if you hit it, the 429 names your limit.

GET https://api.apexapi.dev/v1/uploads
Authorization: Bearer ak-your-api-key

Lists your still-alive uploads (newest first: url, key, bytes, uploaded_at), so a lost URL never forces a re-upload.

List Generations

Your completed generations are reusable as inputs, no re-upload needed. Outputs live on permanent hosted URLs, so any listed url can go straight into image_urls, video_url or an edit source in a new generation request. Listing is free.

Query parameters

GET https://api.apexapi.dev/v1/generations?kind=image&limit=20
Authorization: Bearer ak-your-api-key

{
  "data": [
    {
      "id": "a5933807-c44f-4eec-b25a-15f27a19b9ea",
      "kind": "image",
      "model": "openai/gpt-image-2",
      "prompt": "a cat in a hat",
      "urls": ["https://uploads.apexapi.dev/generated/1a2b.../0.png"],
      "created_at": "2026-07-15T00:00:00.000Z"
    }
  ],
  "next_cursor": "MjAyNi0wNy0xNVQwMD..."
}

Only completed generations with a still-hosted output are listed, so a page can carry fewer rows than limit. Keep following next_cursor until it is null.

Audio (Text-to-Speech)

Generate natural voiceover from text using ElevenLabs models. Returns a URL to an MP3 file. Billed per 1,000 characters.

Parameters

Request

POST https://api.apexapi.dev/v1/audio/speech
Authorization: Bearer ak-your-api-key
Content-Type: application/json

{
  "model": "elevenlabs/tts/multilingual-v2",
  "input": "Welcome to ApexAPI — one API for every AI model.",
  "voice": "Rachel"
}

Response

{
  "created": 1710000000,
  "data": [
    { "url": "https://.../output.mp3" }
  ]
}

List Models

Retrieve a list of all available models and their pricing.

Response

{
  "object": "list",
  "data": [
    {
      "id": "mistralai/mistral-large-2512",
      "object": "model",
      "created": 1782293692,
      "owned_by": "mistralai",
      "type": "chat",
      "display_name": "Mistral: Mistral Large 3 2512",
      "context_length": 262144,
      "max_output_tokens": null,
      "capabilities": {
        "streaming": true,
        "images": true,
        "vision": true,
        "tools": true,
        "image_edit": false,
        "max_reference_images": 0
      },
      "pricing": {
        "input": 6e-07,
        "output": 1.8e-06,
        "unit": "per_token",
        "display": "$0.6/M input, $1.80/M output tokens"
      }
    }
  ]
}

capabilities.vision means the model reads an image you send it in a chat message. capabilities.images is the older name for the same thing and still ships. capabilities.image_edit is the different question: whether an image model accepts a reference image you pass to /v1/images/generations, and capabilities.max_reference_images is how many it takes. Pricing reflects the rail that will actually serve your request, so the number you see is the number you pay.

Context for AI (web data)

Give models live web knowledge on the same key and balance. Billed per unit: crawl and extract failures are free, and scrape is billed once the fetch is dispatched to a rail (a blocked page or unreadable content is still billed; a rejected URL is not). Full guides: read a page, read a site, structured extract.

curl https://api.apexapi.dev/v1/scrape \
  -H "Authorization: Bearer ak-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{ "url": "https://example.com/pricing", "stealth": "auto" }'
# → { "content": "# Pricing…", "tier": "standard", "cost": 0.002, … }

curl https://api.apexapi.dev/v1/crawl \
  -H "Authorization: Bearer ak-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{ "url": "https://docs.example.com", "limit": 200 }'
# → { "id": "…", "status": "running" }  then poll GET /v1/crawl/{id}

curl https://api.apexapi.dev/v1/extract \
  -H "Authorization: Bearer ak-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{ "scraper": "amazon-product",
        "input": { "url": "https://www.amazon.com/dp/B09B8V1LZ3" } }'
# → { "id": "…", "status": "running" }  then poll GET /v1/extract/{id}
# List scrapers: GET /v1/extract/scrapers