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

# Models and packages

> Which model belongs to which package, how to choose, how to verify

## Two packages, two models

| Package | Model name in your app | Use it for |
| - | - | - |
| **GPT** | `gpt-5.6-sol` | Coding, refactoring, debugging, multi-step reasoning, long documents. The standard package, cheapest per token |
| **GPT Astra** | `gpt-astra` | The newest generation, stronger than `sol`, higher price per token. For the hardest work |
| **Claude** | `claude-opus-5`, `claude-opus-5-5`, `claude-fable-5-1`, `claude-fable-5`, `claude-sonnet-5`, `claude-sonnet-4-6`, `claude-haiku-4-5` | Claude Code, the Anthropic SDK or any OpenAI‑compatible app. Credits per model: Opus ≈ 1/token, Fable ≈ 2, Sonnet/Haiku cheaper |

The model follows the package: **a GPT key can only call `gpt-5.6-sol`, an Astra key can only call `gpt-astra`**. Calling across packages returns "No available channel for model ...". To use both, buy two codes and configure one key per app.

<Warning>
  Model names must match **exactly**, case-sensitive. `gpt-5.6-Sol`, `gpt5.6-sol`, `GPT-astra` are all wrong. Copy from the table.
</Warning>

## Shared capabilities

* **Chat Completions** and **Responses API**, with streaming, tool calling and JSON mode.
* **Vision**: send screenshots of errors, charts, images with text. See [cURL / Python](/en/setup/api).
* **Reasoning**: both are reasoning models. You may set `reasoning_effort` (`low`, `medium`, `high`) in the request; if you do not, the gateway uses a high level for consistent quality. Reasoning tokens are reported in `usage.completion_tokens_details.reasoning_tokens` and count against quota like output tokens.

## How to choose

<Tabs>
  <Tab title="Coding">
    `gpt-5.6-sol` handles most coding work. For very hard tasks, deep reasoning or very long context, `gpt-astra`.
  </Tab>

  <Tab title="Translation, summaries, chat">
    `gpt-5.6-sol`. Mechanical work does not need the expensive model; to save further add `"reasoning_effort": "low"` when calling from code.
  </Tab>

  <Tab title="Long-running agents (Codex, Cline)">
    Agents resend the whole context every turn, so tokens go fast. `gpt-5.6-sol` is the balanced choice; switch to `gpt-astra` only for tasks `sol` cannot finish.
  </Tab>
</Tabs>

## List the models your key can call

```bash theme={null}
curl https://gpt.teamsoclo.site/v1/models \
  -H "Authorization: Bearer sk-xxx"
```

The result lists exactly the models this key is allowed to use. Anything not listed cannot be called with it.

## Quota and rate

* Quota is charged by tokens actually used (input + output + reasoning), exactly the `usage` returned.
* Each key has a per-minute request limit set by its package; exceeding it returns `429`, wait for the next minute. Add a delay between calls in batch scripts.
* Remaining quota, expiry date and allowed models: [gpt.teamsoclo.site/check](https://gpt.teamsoclo.site/check).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.