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

# Models

> Available models for the BLACKBOX Agent API. Use these model IDs in the `model` field when creating tasks.

The Agent API supports two execution modes — **Claude Agent** (default) and **Standard Chat** — each with their own set of available models.

## Agent Runtimes — Claude vs Codex

Within the Claude Agent mode (`type: "claude"`), the API runs your task on one of **two agent runtimes**, selected automatically from the `model` id:

| Runtime                    | SDK                              | Wire protocol                       | Used for                                     |
| -------------------------- | -------------------------------- | ----------------------------------- | -------------------------------------------- |
| **Claude Agent** (default) | `@anthropic-ai/claude-agent-sdk` | Anthropic Messages (`/v1/messages`) | Anthropic/Claude ids and all other providers |
| **Codex Agent**            | `@openai/codex-sdk`              | OpenAI Responses (`/v1/responses`)  | OpenAI/Codex ids                             |

**Selection rule** — a model id runs on the **Codex** runtime when it contains `codex`, `openai/`, `gpt-`, or starts with `o1`/`o3`/`o4`; otherwise it runs on the **Claude** runtime.

| Example model id                                                           | Runtime |
| -------------------------------------------------------------------------- | ------- |
| `blackboxai/openai/gpt-5.3-codex`                                          | Codex   |
| `blackboxai/openai/gpt-5.5` (and `gpt-5.4`, `gpt-5.4-pro`, `gpt-5.4-nano`) | Codex   |
| `blackboxai/anthropic/claude-opus-4.6` / `claude-sonnet-4.5`               | Claude  |
| everything else (nemotron, minimax, glm, …)                                | Claude  |

Both runtimes are fully managed — they run inside the same sandbox, expose the **same task lifecycle and event stream**, and are interchangeable from the caller's perspective. You normally don't pick the runtime directly; you pick the `model` and the runtime follows.

<Note>
  **Explicit override** — pass `agent: "claude"` or `agent: "codex"` to force a runtime regardless of the model id (e.g. `agent: "codex"` with a non-OpenAI model). Omit `agent` to auto-select from the model id (default: **claude**). The benchmark endpoint (`POST /api/v1/benchmarks/runs`) accepts the same `agent` field.
</Note>

## Claude Agent Models

Used when `type: "claude"` (the default). Pass these IDs in the `model` field of your request body. The **agent runtime** (Claude vs Codex) is auto-selected per the rule above.

| Model Name                    | Model ID                                         | Notes                                              |
| ----------------------------- | ------------------------------------------------ | -------------------------------------------------- |
| Claude Sonnet 4.6             | `blackboxai/anthropic/claude-sonnet-4.6`         | **Default** — best balance of speed and capability |
| Blackbox Pro                  | `blackboxai/blackbox-pro`                        | Blackbox flagship agent model                      |
| Claude Opus 4.7               | `blackboxai/anthropic/claude-opus-4.7`           | Latest Opus generation                             |
| Claude Opus 4.6               | `blackboxai/anthropic/claude-opus-4.6`           | Most capable, slower                               |
| Claude Sonnet 4.5             | `blackboxai/anthropic/claude-sonnet-4.5`         | Previous Sonnet generation                         |
| Claude Haiku 4.5              | `blackboxai/anthropic/claude-haiku-4.5`          | Fast, lightweight Claude model                     |
| GPT-5.5 Pro                   | `blackboxai/openai/gpt-5.5-pro`                  | OpenAI top-tier Pro model                          |
| GPT-5.5                       | `blackboxai/openai/gpt-5.5`                      | OpenAI flagship                                    |
| GPT-5.4 Pro                   | `blackboxai/openai/gpt-5.4-pro`                  | OpenAI high-capability model                       |
| GPT-5.4                       | `blackboxai/openai/gpt-5.4`                      | OpenAI general-purpose model                       |
| GPT-5.4 Nano                  | `blackboxai/openai/gpt-5.4-nano`                 | Compact, low-latency OpenAI model                  |
| GPT-5.3 Codex                 | `blackboxai/openai/gpt-5.3-codex`                | OpenAI code-specialized model                      |
| Nemotron 3 Nano 30B A3B       | `blackboxai/nvidia/nemotron-3-nano-30b-a3b`      | NVIDIA Nemotron compact MoE                        |
| Nemotron Nano 12B v2 VL       | `blackboxai/nvidia/nemotron-nano-12b-v2-vl`      | NVIDIA Nemotron vision-language                    |
| Grok 4.3                      | `blackboxai/x-ai/grok-4.3`                       | xAI flagship                                       |
| Grok 4.1 Fast (Non-Reasoning) | `blackboxai/x-ai/grok-4.1-fast-non-reasoning`    | xAI fast non-reasoning model                       |
| Grok Build 0.1                | `blackboxai/x-ai/grok-build-0.1`                 | xAI build-focused model                            |
| MiniMax M2.7                  | `blackboxai/minimax/minimax-m2.7`                | Latest MiniMax flagship                            |
| MiniMax M2.5                  | `blackboxai/minimax/minimax-m2.5`                | MiniMax previous flagship                          |
| Gemini 3.5 Flash              | `blackboxai/google/gemini-3.5-flash`             | Google Gemini fast multimodal                      |
| Gemini 3.1 Pro Preview        | `blackboxai/google/gemini-3.1-pro-preview`       | Google Gemini Pro preview                          |
| Gemini 3.1 Flash Lite         | `blackboxai/google/gemini-3.1-flash-lite`        | Google Gemini lightweight                          |
| Gemma 4 26B A4B IT            | `blackboxai/google/gemma-4-26b-a4b-it`           | Google Gemma instruction-tuned MoE                 |
| Kimi K2.6                     | `blackboxai/moonshotai/kimi-k2.6`                | Moonshot AI flagship                               |
| Devstral 2                    | `blackboxai/mistral/devstral-2`                  | Mistral developer model                            |
| Ministral 8B                  | `blackboxai/mistral/ministral-8b`                | Mistral compact 8B model                           |
| Ministral 3B                  | `blackboxai/mistral/ministral-3b`                | Mistral compact 3B model                           |
| Pixtral Large                 | `blackboxai/mistral/pixtral-large`               | Mistral large multimodal model                     |
| Pixtral 12B                   | `blackboxai/mistral/pixtral-12b`                 | Mistral multimodal model                           |
| Codestral                     | `blackboxai/mistral/codestral`                   | Mistral code-specialized model                     |
| Mistral Medium 3.5            | `blackboxai/mistral/mistral-medium-3.5`          | Mistral medium-tier model                          |
| Mistral Small                 | `blackboxai/mistral/mistral-small`               | Mistral small general-purpose model                |
| Mistral Nemo                  | `blackboxai/mistral/mistral-nemo`                | Mistral Nemo model                                 |
| Llama 3.2 3B                  | `blackboxai/meta/llama-3.2-3b`                   | Meta Llama compact                                 |
| Llama 3.2 1B                  | `blackboxai/meta/llama-3.2-1b`                   | Meta Llama smallest                                |
| Llama 3.1 70B                 | `blackboxai/meta/llama-3.1-70b`                  | Meta Llama 3.1 large                               |
| Llama 3.1 8B                  | `blackboxai/meta/llama-3.1-8b`                   | Meta Llama 3.1                                     |
| Morph v3 Large                | `blackboxai/morph/morph-v3-large`                | Morph large model                                  |
| Morph v3 Fast                 | `blackboxai/morph/morph-v3-fast`                 | Morph fast model                                   |
| Nova 2 Lite                   | `blackboxai/amazon/nova-2-lite`                  | Amazon Nova lightweight model                      |
| Nova Micro                    | `blackboxai/amazon/nova-micro`                   | Amazon Nova smallest model                         |
| DeepSeek V4 Flash             | `blackboxai/deepseek/deepseek-v4-flash`          | DeepSeek fast model                                |
| MiMo v2.5 Pro                 | `blackboxai/xiaomi/mimo-v2.5-pro`                | Xiaomi MiMo flagship                               |
| MiMo v2.5                     | `blackboxai/xiaomi/mimo-v2.5`                    | Xiaomi MiMo base model                             |
| Qwen 3.7 Max                  | `blackboxai/alibaba/qwen3.7-max`                 | Alibaba Qwen next-gen flagship                     |
| Qwen 3.6 Max Preview          | `blackboxai/alibaba/qwen3.6-max-preview`         | Alibaba Qwen flagship preview                      |
| Qwen 3.6 Plus                 | `blackboxai/alibaba/qwen3.6-plus`                | Alibaba Qwen Plus tier                             |
| Qwen 3.6 27B                  | `blackboxai/alibaba/qwen3.6-27b`                 | Alibaba Qwen mid-size                              |
| Qwen3 VL 235B A22B Instruct   | `blackboxai/alibaba/qwen3-vl-235b-a22b-instruct` | Qwen3 vision-language MoE                          |
| Qwen3 Max Thinking            | `blackboxai/alibaba/qwen3-max-thinking`          | Qwen3 reasoning-focused                            |
| Qwen3 Max                     | `blackboxai/qwen3-max`                           | Qwen3 flagship                                     |
| Qwen3 Coder                   | `blackboxai/qwen3-coder`                         | Qwen3 code-specialized                             |
| Qwen3 VL 235B A22B            | `blackboxai/qwen3-vl-235b-a22b`                  | Qwen3 vision-language MoE                          |
| Qwen3 VL 32B                  | `blackboxai/qwen3-vl-32b`                        | Qwen3 vision-language 32B                          |
| GLM-5                         | `blackboxai/z-ai/glm-5`                          | Z.AI GLM flagship                                  |
| GLM-5.1                       | `blackboxai/z-ai/glm-5.1`                        | Z.AI GLM next iteration                            |
| GLM-5V Turbo                  | `blackboxai/z-ai/glm-5v-turbo`                   | Z.AI GLM vision turbo                              |
| GLM-4.7 Flash                 | `blackboxai/z-ai/glm-4.7-flash`                  | Z.AI GLM fast model                                |
| Trinity Large Thinking        | `blackboxai/arcee-ai/trinity-large-thinking`     | Arcee reasoning model                              |
| KAT Coder Pro v2              | `blackboxai/kwaipilot/kat-coder-pro-v2`          | Kwaipilot code model                               |

<Note>
  If you omit the `model` field, the API defaults to `blackboxai/anthropic/claude-sonnet-4.6`.
</Note>

### Example — Single Agent with Model Override

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
curl -X POST 'https://agent.blackbox.ai/api/v1/tasks' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "prompt": "Refactor the auth module to use middleware",
    "model": "blackboxai/anthropic/claude-opus-4.7"
  }'
```

***

## Standard Chat Models

Used when `type: "standard"`. Pass these IDs in the `model` field.

| Model Name    | Model ID                          | Notes                                   |
| ------------- | --------------------------------- | --------------------------------------- |
| Mistral Small | `mistral/mistral-small`           | **Default** — fast, streaming, tool use |
| MiniMax M2.5  | `minimax/minimax-m2.5`            | MiniMax flagship                        |
| Kimi K2.5     | `moonshotai/kimi-k2.5`            | Moonshot AI flagship                    |
| Grok 4.1 Fast | `xai/grok-4.1-fast-non-reasoning` | xAI fast non-reasoning model            |

<Note>
  If you omit the `model` field in standard mode, the API defaults to `mistral/mistral-small`.
</Note>

### Example — Standard Chat with Model Override

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
curl -X POST 'https://agent.blackbox.ai/api/v1/tasks' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "prompt": "Explain async/await in JavaScript",
    "type": "standard",
    "model": "moonshotai/kimi-k2.5"
  }'
```