> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prisminference.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Models

> Model IDs, context windows, and workload guidance for Prism inference.

Use the model ID in the `model` field of either supported API format. All
models are available through `api.prisminference.com`.

| Model             | Model ID                  | Context | Use when                                                          |
| ----------------- | ------------------------- | ------- | ----------------------------------------------------------------- |
| GLM-5.3           | `prism/glm-5.3`           | 1M      | An agent needs long context, tool use, and multi-step coding.     |
| Kimi K3           | `prism/kimi-k3`           | 1M      | A task needs deep reasoning, vision, or a long conversation.      |
| Qwen              | `prism/qwen`              | 256K    | Inner-loop work benefits from a smaller, low-latency dense model. |
| DeepSeek-V4-Flash | `prism/deepseek-v4-flash` | 1M      | Coding and tool-use loops prioritize fast generation.             |

<h2 id="glm-53">
  GLM-5.3
</h2>

`prism/glm-5.3` is the default model for coding agents. It is a 744B sparse
mixture-of-experts model with a 1M-token context window.

```json theme={"theme":{"light":"github-light","dark":"github-dark"}}
{
  "model": "prism/glm-5.3",
  "messages": [
    { "role": "user", "content": "Review this repository-wide refactor: ..." }
  ]
}
```

<h2 id="kimi-k3">
  Kimi K3
</h2>

`prism/kimi-k3` is a 2.8T mixture-of-experts model with roughly 50B active
parameters and a 1M-token context window. Choose it for reasoning-heavy coding
and multimodal input.

<h2 id="qwen">
  Qwen
</h2>

`prism/qwen` is a 27B dense model with a 256K-token context window. Choose it
for short, frequent agent turns where latency matters more than maximum model
capacity.

<h2 id="deepseek-v4-flash">
  DeepSeek-V4-Flash
</h2>

`prism/deepseek-v4-flash` is the fast model in the public lineup. It has a 1M-token
context window and is intended for coding, reasoning, and tool use.

## Shared interface

The public models use the same request formats:

* OpenAI Chat Completions at `POST /v1/chat/completions`
* Anthropic Messages at `POST /v1/messages`
* Streaming in either format
* OpenAI function tools or Anthropic tool-use blocks
* JSON mode and JSON Schema structured output on Chat Completions
* Optional reasoning controls

<Tip>
  Start with `prism/glm-5.3`. Move to a different model when a measured workload
  shows a clear need for lower latency, multimodal input, or a different
  reasoning profile.
</Tip>

## List available models

Use the public Models API to read the current catalog. No API key is required:

```bash theme={"theme":{"light":"github-light","dark":"github-dark"}}
curl "https://api.prisminference.com/models"
```

```json theme={"theme":{"light":"github-light","dark":"github-dark"}}
{
  "object": "list",
  "data": [
    {
      "id": "prism/glm-5.3",
      "object": "model",
      "created": 1788926400,
      "owned_by": "prism"
    }
  ]
}
```

Retrieve one model by ID:

```bash theme={"theme":{"light":"github-light","dark":"github-dark"}}
curl "https://api.prisminference.com/v1/models/prism%2Fglm-5.3"
```

`/v1/models` is an OpenAI-compatible alias for `/models`. The endpoint returns
`404` when a model does not exist. Use the list endpoint at runtime instead of
assuming every documented model is still available.