Models and capabilities

What does BeatAPI serve?

GET /v1/models returns the text models enabled for your account. The pricing page lists everything with its current rate.

Which request formats work?

Four, and every text model answers on all of them:

Choose whichever your existing code already speaks. A Claude model over the OpenAI format and a GPT model over the Anthropic format both work — the format is the wire shape, not a restriction on which models you can reach.

What is deliberately not served?

/v1/embeddings, /v1/rerank, /v1/moderations, and /v1/audio/* return 404. They are not planned-but-missing; they are not part of the published surface.

The practical consequence: a tool that wants an embedding model alongside a chat model — AnythingLLM is the common case — needs its embedder pointed somewhere else, usually the local one it ships with.

Is image and video generation OpenAI-compatible?

No, and deliberately. BeatAPI’s generation API is asynchronous and task-based:

POST /v1/images/tasks   ->  201, { "data": { "id": "..." } }
GET  /v1/tasks/{id}     ->  poll until succeeded or failed

There is no synchronous /v1/images/generations. A model that takes minutes cannot answer inside one HTTP request without either lying about progress or timing out, so the task ID is returned immediately and the result is fetched when it exists. Failed tasks are refunded, which a synchronous call has no clean way to express.

See Development Guide for the lifecycle, and Webhooks to avoid polling.

Why does the API answer differently from a chat product?

A consumer chat product is an application built around a model. The API gives you the model.

What the product adds, and you do not get for free:

• Web search and other live tools
• Conversation memory across sessions
• A code execution sandbox
• A tuned system prompt
• Retrieval over your uploaded files

Through the API you manage conversation history yourself, set your own system prompt, and wire up tool calling if you want it. Nothing is missing — the assembly is yours.

Do models support images, tools, and streaming?

Streaming works on every text model via stream: true.

Vision and tool calling vary by model. deepseek-v4-flash-vision-exp accepts image input and is priced identically to the text-only Flash tier; the Claude, GPT, and Gemini families are broadly multimodal. Check the model’s page under Text API before designing around a capability.

How do I choose a model?

Model IDs are stable. When a family gains a generation it gets a new ID; an existing ID is not repointed at different weights underneath you.

Do prices change with time or context length?

For some models, yes, and both are published:

• Peak hours — the DeepSeek family doubles during weekday peak hours in Beijing time. Hunyuan hy3 has its own daily window.
• Context tiers — the GPT-5.6 and Grok families bill a higher rate past a context threshold.

Exact hours and thresholds are in the pricing page footnotes. Every response’s usage record shows which rate was applied.

Can I use the models I already pay for elsewhere?

BeatAPI is a single account and a single balance. There is no arrangement for attaching a key you hold with another vendor.