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

# Model Roster

> Public virtual models and hosted model roles available through Astrolabe Cloud.

Astrolabe Cloud exposes stable virtual model IDs while the hosted roster can evolve behind the gateway.

Use virtual models for most traffic. Request a concrete model only when you want to bypass virtual lane selection while still keeping Cloud auth, billing, stack policy, and traces.

## Virtual models

| Model                   | Purpose                                                                 |
| ----------------------- | ----------------------------------------------------------------------- |
| `astrolabe/auto`        | Default route for general AI API requests.                              |
| `astrolabe/coding`      | Code generation, patching, repo work, and agent execution traces.       |
| `astrolabe/research`    | Long-context synthesis, source-heavy research, and deep reasoning.      |
| `astrolabe/vision`      | Screenshots, documents, image-grounded reasoning, and multimodal tasks. |
| `astrolabe/strict-json` | Structured output and schema recovery.                                  |
| `astrolabe/cheap`       | Low-risk, high-efficiency turns.                                        |
| `astrolabe/safe`        | High-stakes routes with conservative premium defaults.                  |

Short aliases such as `auto`, `coding`, and `cheap` normalize to their `astrolabe/...` model IDs.

## Active hosted roster

Astrolabe Cloud maintains a hosted pricing catalog. The current public serving roster includes:

| Model ID                       | Role                                                |
| ------------------------------ | --------------------------------------------------- |
| `deepseek/deepseek-v4-pro`     | Workhorse for `auto`, `coding`, and `research`.     |
| `minimax/minimax-m2.7`         | Conservative workhorse and safety fallback.         |
| `xiaomi/mimo-v2.5-pro`         | Coding and research specialist.                     |
| `qwen/qwen3.6-plus`            | Long-context research and vision candidate.         |
| `moonshotai/kimi-k2.6`         | Multimodal and visual coding specialist.            |
| `anthropic/claude-sonnet-4.6`  | Safe-lane default and premium escalation.           |
| `z-ai/glm-5.1`                 | Strict JSON and schema repair specialist.           |
| `deepseek/deepseek-v4-flash`   | Low-cost coding and structured-output fallback.     |
| `mistralai/mistral-small-2603` | Value multimodal fallback.                          |
| `x-ai/grok-4.1-fast`           | Low-cost long-context and tool-capable model.       |
| `x-ai/grok-4.3`                | Premium research and agentic candidate.             |
| `openai/gpt-5.4`               | Premium verifier or executor.                       |
| `anthropic/claude-opus-4.7`    | Safe-lane escalation.                               |
| `openai/gpt-5.5`               | Frontier escalation.                                |
| `google/gemma-4-31b-it`        | Cheap multimodal generalist.                        |
| `openai/gpt-5.4-nano`          | Control-plane, router, and structured-output model. |
| `openai/gpt-5.4-mini`          | Mid-premium structured and safe-lane option.        |

## Pricing

Retail prices are returned by `GET /v1/models` and reflected in the Cloud app pricing views. Billing uses provider-confirmed usage where available.

Routing model costs are internal to Astrolabe. Customer debits are based on serving or fallback model usage that is billable.

## Choosing models

Use:

* `astrolabe/auto` for most traffic
* `astrolabe/coding` for code editing and tool loops
* `astrolabe/research` for synthesis and long-context analysis
* `astrolabe/vision` for visual or document-grounded requests
* `astrolabe/strict-json` for schema-sensitive requests
* `astrolabe/cheap` for low-risk volume
* `astrolabe/safe` for sensitive or high-stakes flows

Use routing stacks to further constrain providers, models, cost mode, quality target, and verification.