One API.
Every intelligence.

Build once. Switch intelligence anytime.

One endpoint. Hundreds of models.

Every provider below is live in the catalog right now, synced automatically — not a fixed list we update by hand. 117 models, one request shape.

OpenAI20

AAlibaba12

Google10

XxAI9

Anthropic7

DeepSeek6

ByteDance5

Cohere4

ZZ.ai4

Xiaomi3

KKling3

BBlack Forest Labs2

• OpenAI20
• AAlibaba12
• Google10
• XxAI9
• Anthropic7
• DeepSeek6
• ByteDance5
• Cohere4
• ZZ.ai4
• Xiaomi3
• KKling3
• BBlack Forest Labs2

+7 more providers in the full model directory.

Change the model. Keep your code.

Pick a different model below and watch the request: one field changes, everything else — messages, headers, parsing — stays exactly as it was.

1{2  "model": "gpt-6-astra",3  "messages": [4    {5      "role": "user",6      "content": "Draft a launch announcement for our new feature."7    }8  ]9}

Build a request. Know its cost before you send it.

Pick a live model and shape the request. You see the code that goes over the wire, the most it can cost at today's API prices, and how the API will treat each parameter. Then run it for real in the developer playground.

1curl https://api.ultragpt.pro/v1/chat/completions \2  -H "Authorization: Bearer $ULTRAGPT_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "gpt-6-astra",6    "messages": [7      {8        "role": "user",9        "content": "Write a short product description for an AI developer platform."10      }11    ],12    "stream": true,13    "stream_options": {14      "include_usage": true15    },16    "temperature": 0.7,17    "max_tokens": 51218  }'

One API. Every format.

The same /v1/messages request carries text and vision. Dedicated endpoints cover image and audio. Coverage below is counted live from the catalog, not promised in copy.

Responses that arrive as they happen.

Add "stream": true and read the response as server-sent events — each event a partial delta, the final one carrying the model id and stop reason.

1curl https://api.ultragpt.pro/v1/chat/completions \2  -H "Authorization: Bearer $ULTRAGPT_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "gpt-6-astra",6    "messages": [7      {8        "role": "user",9        "content": "What's next for AI infrastructure?"10      }11    ],12    "stream": true,13    "stream_options": {14      "include_usage": true15    }16  }'

Get JSON back. Call your own tools.

{
  "product": "UltraGPT API",
  "price_usd": 19.9,
  "features": [
    "one request shape",
    "400+ models",
    "streaming"
  ]
}

From zero to production in minutes.

1. 01

   Create your API key

   Sign in to the app, open Settings → API keys, and generate a key. Store it as an environment variable — never in client-side code.

2. 02

   Choose a model

   Pick any model id from the catalog — a fast everyday chat model or a heavier reasoning one. The request shape never changes.

3. 03

   Send your first request

   POST to /v1/messages with your key and a messages array. Swap languages below; the request is identical either way.

1curl https://api.ultragpt.pro/v1/chat/completions \2  -H "Authorization: Bearer $ULTRAGPT_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "gpt-6-astra",6    "messages": [7      {8        "role": "user",9        "content": "Hello!"10      }11    ]12  }'

Need a key first? Create one in the app →

Reference for everything above.

Every endpoint and parameter shown on this page has a full write-up in the real docs.

Built for applications, not experiments.

Built to hold up under real traffic.

The dashboard below is the shape a status page takes once it's wired to live metrics — the numbers shown now are an example, not a report.

Simple usage-based API access.

API access runs through your UltraGPT subscription — the same plan as the app, no separate per-token bill to manage. Rates below are the same per-model figures shown in the model directory, useful for estimating what a given model costs to run.

Web search: $0.008 / search · Bring your own key: 5% fee · Cached input bills at each model's lower cached rate · Response-cache hits are free.

Built for what you're building.

Start with one request. Scale to millions.

The same request shape carries a weekend prototype and a production integration — nothing to migrate when one becomes the other.

Questions developers ask.