> ## Documentation Index
> Fetch the complete documentation index at: https://platform.minimax.io/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Codex
> Use the latest MiniMax M-series models for AI programming in the Codex desktop app.
[**Codex**](https://developers.openai.com/codex/) is OpenAI's official AI coding agent for the desktop.
## Install Codex
You can skip Codex CLI installation when using the one-click setup wizard below. If it cannot find the `codex` command, it offers to install the official `@openai/codex` npm package.
Download and install the Codex desktop app from the [OpenAI Codex page](https://developers.openai.com/codex/).
## Configure MiniMax API
Run:
```bash theme={null}
npx -y mmx-cli@latest agent setup
```
Select **Codex**. If Codex CLI is missing, the wizard offers to install it. **Codex CLI 0.146.0 or newer** is recommended. The wizard configures `~/.codex/config.toml` and `~/.codex/mmx-model-catalog.json`.
Close Codex and any editor that has `~/.codex/config.toml` open before running the wizard. This prevents another process from saving stale contents over the configuration it just wrote.
Restart Codex after setup. See [MiniMax CLI](/docs/m-plan/minimax-cli) for command options and backup behavior.
Open `~/.codex/config.toml` and add or update the following entries, replacing `` with your key from the [MiniMax Developer Platform](https://platform.minimax.io/console/plan):
```toml theme={null}
model = "MiniMax-M3.1-Flash-Preview"
model_provider = "minimax"
model_context_window = 524288
preferred_auth_method = "apikey"
forced_login_method = "api"
[model_providers.minimax]
name = "MiniMax"
base_url = "https://api.minimax.io/v1"
experimental_bearer_token = ""
wire_api = "responses"
```
`preferred_auth_method` and `forced_login_method` select API Key authentication. To show model capabilities and reasoning levels in `/model`, you can also add the optional catalog below.
Restart Codex to start using MiniMax-M3.1-Flash-Preview.
### Configure Model Capabilities (Optional)
For manual setup, you can add a model capability catalog so Codex shows MiniMax-M3.1-Flash-Preview and its reasoning levels in `/model`.
Add this line to `~/.codex/config.toml`:
```toml theme={null}
model_catalog_json = "~/.codex/model-catalogs/custom-catalog.json"
```
Then create `~/.codex/model-catalogs/custom-catalog.json` with the model configuration:
```json theme={null}
{
"models": [
{
"slug": "MiniMax-M3.1-Flash-Preview",
"display_name": "MiniMax-M3.1-Flash-Preview",
"description": "MiniMax M3.1 Flash Preview",
"default_reasoning_level": "max",
"supported_reasoning_levels": [
{ "effort": "low", "description": "Low" },
{ "effort": "medium", "description": "Medium" },
{ "effort": "high", "description": "High" },
{ "effort": "xhigh", "description": "Extra high" },
{ "effort": "max", "description": "Maximum" }
],
"shell_type": "shell_command",
"visibility": "list",
"supported_in_api": true,
"priority": 0,
"base_instructions": "You are Codex, a coding agent based on MiniMax-M3.1-Flash-Preview. You and the user share the same workspace and collaborate to achieve the user's goals.",
"supports_reasoning_summaries": true,
"default_reasoning_summary": "none",
"support_verbosity": false,
"truncation_policy": { "mode": "tokens", "limit": 10000 },
"supports_parallel_tool_calls": true,
"experimental_supported_tools": [],
"prefer_websockets": false,
"apply_patch_tool_type": "freeform",
"web_search_tool_type": "text",
"supports_image_detail_original": false,
"tool_mode": "code_mode_only",
"multi_agent_version": "v2",
"use_responses_lite": false,
"input_modalities": ["text", "image"],
"context_window": 524288,
"max_context_window": 524288,
"effective_context_window_percent": 95,
"auto_compact_token_limit": null,
"reasoning_summary_format": "experimental",
"supports_search_tool": true
}
]
}
```
Do not add `{ "effort": "none" }` to `supported_reasoning_levels`. MiniMax-M3.1-Flash-Preview always thinks, so selecting that level makes Codex send `reasoning: {"effort": "none"}` and the API returns `400`.
Common fields in this configuration:
* `slug` / `display_name`: The model identifier and display name used in Codex config and the `/model` list. Keep this aligned with the model name used by the API.
* `default_reasoning_level`: The default reasoning effort. MiniMax-M3.1-Flash-Preview uses `max` by default.
* `supported_reasoning_levels`: Reasoning options users can switch between in `/model`. MiniMax-M3.1-Flash-Preview supports `low`, `medium`, `high`, `xhigh`, and `max`; `none` is not supported.
* `base_instructions`: Base system prompt Codex adds when using this model. Use it to describe the model identity and collaboration style.
* `supports_reasoning_summaries`: Enables Codex's Responses API reasoning path for this model. Set this to `true` so Codex sends `reasoning.effort`; otherwise Codex omits the `reasoning` field even when `default_reasoning_level` is configured. In this example, `default_reasoning_summary` is set to `none` so Codex does not request a separate reasoning summary.
* `shell_type`: Declares the shell tool-call type supported by the model. This example uses `shell_command`.
* `visibility` / `supported_in_api` / `priority`: Control whether the model appears in the list, whether it is available through the API, and its ordering priority in the model list.
* `supports_parallel_tool_calls`: Indicates that the model supports parallel tool calls, allowing Codex to handle multiple tool requests.
* `experimental_supported_tools`: Reserved list for experimental tool capabilities. Leave it as an empty array when no extra tools are needed.
* `input_modalities`: Supported input modalities. `["text", "image"]` means text and image input are supported.
* `truncation_policy`: Controls content truncation; this example counts tokens.
* `context_window` / `max_context_window`: Set the client context window to the recommended 512K (524,288 tokens) in this example. See the [512K guidance](#context-window-512k).
* `effective_context_window_percent` / `auto_compact_token_limit`: Use 95% of the window as the effective context budget without a separate fixed auto-compaction threshold.
Restart Codex after changing the model catalog.
### Context Window and Cost Optimization (Optional)
MiniMax-M3.1-Flash-Preview supports a context window of up to 1M tokens. For everyday coding and other multi-turn tasks, consider setting your tool's context window to **512K**. Managing and compacting conversation history earlier can reduce the number of historical tokens carried into subsequent requests. Use **1M** when your task requires retaining extensive code or document details throughout the conversation.
#### Why Consider 512K?
In a ProgramBench comparison using MiniMax-M3.1-Flash-Preview, average model performance was similar with 512K and 1M context windows, while the estimated cost with 512K was approximately **20% lower**. Consider 512K as an optional configuration for everyday coding.
**Model Performance and Usage Across Context Windows**
| Metric | 1M window | 512K window | 512K vs. 1M |
| - | - | - | - |
| Evaluation tasks | 200 | 200 | Same |
| Model performance | Similar | Similar | No clear difference detected |
| Average total tokens per task | 136.76 million | 111.42 million | **18.5% fewer** |
| Estimated usage-based charges | 100% | 80.1% | **19.9% lower** |