MiniMax M3.1-Flash Codex Manual Setup and Model Catalog Configuration

One-sentence takeaway

The official Codex integration page's Manual setup uses the Responses API and writes MiniMax-M3.1-Flash to ~/.codex/config.toml; the optional model catalog exposes low, medium, high, xhigh, and max to /model, and explicitly prohibits none.

Use cases

• Suitable tasks: Using MiniMax M3.1-Flash in Codex for terminal programming and multi-turn Agent work.

• Unsuitable tasks: Scenarios that require thinking to be disabled, or that require treating manual editing as an automatic configuration backup.

• Applicable model version: MiniMax-M3.1-Flash.

• Applicable client, Agent, or API: Codex CLI/desktop app; MiniMax Responses API.

• Recommended reasoning levels and parameters: The model catalog defaults to max and supports low, medium, high, xhigh, and max; none is unsupported. The page recommends using a 524288 (512K) context for everyday coding.

Ready-to-use content

~/.codex/config.toml

Open ~/.codex/config.toml and replace <MINIMAX_API_KEY> with your MiniMax key. The following is the official Manual setup configuration:

model = "MiniMax-M3.1-Flash"
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 = "<MINIMAX_API_KEY>"
wire_api = "responses"

Model catalog (optional)

Add the following field at the top level of ~/.codex/config.toml, before any [model_providers.*] table:

model_catalog_json = "~/.codex/model-catalogs/custom-catalog.json"

Create ~/.codex/model-catalogs/custom-catalog.json:

{
  "models": [
    {
      "slug": "MiniMax-M3.1-Flash",
      "display_name": "MiniMax-M3.1-Flash",
      "description": "MiniMax M3.1 Flash",
      "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. 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
    }
  ]
}

Reasoning levels that cannot be added

Do not add the following value to supported_reasoning_levels:

{ "effort": "none" }

The official documentation states that M3.1-Flash always reasons; selecting none causes Codex to send reasoning: {"effort": "none"}, and the API returns 400.

Test or workflow steps

1. Install or open Codex. The page recommends Codex CLI 0.146.0 or later.

2. Edit ~/.codex/config.toml and add the provider configuration above.

3. If you want the model and reasoning levels to appear in /model, create the model catalog file and set model_catalog_json in config.toml.

4. Restart Codex after modifying the model catalog.

5. In /model, select MiniMax-M3.1-Flash and an available reasoning level.

Original evidence and data

• The API model ID, catalog slug, and display_name are all MiniMax-M3.1-Flash.

• The provider uses base_url = "https://api.minimax.io/v1" and wire_api = "responses"; the authentication method is specified by preferred_auth_method = "apikey" and forced_login_method = "api".

• The official catalog sets the default reasoning level to max and supports low, medium, high, xhigh, and max.

• The catalog sets the context window to 524288 and the effective context percentage to 95, while retaining a note that the maximum context is 1M. The page cites a comparison of 200 ProgramBench tasks: the average total tokens were 111.42M at 512K and 136.76M at 1M; estimated usage costs were about 19.9% lower, with similar model performance. This is the official comparison disclosed on the page, not an independent re-test by this article.

• The one-click wizard command is npx -y mmx-cli@latest agent setup; when Codex is selected, it configures ~/.codex/config.toml and ~/.codex/mmx-model-catalog.json.

• Before starting the wizard, close Codex and any editor with ~/.codex/config.toml open, to prevent another process from saving old content back over the new configuration; restart Codex after configuration is complete.

• The official MiniMax CLI documentation states that the wizard creates a timestamped .bak before modifying existing files, restores modified files on failure, and preserves other providers and unrelated settings. This backup behavior applies to the wizard; the manual setup page does not promise automatic backups.

Scope and limitations

• This is the Codex configuration for MiniMax M Plan; its fields cannot be assumed to apply to other clients or other MiniMax models.

• 1M is the capability limit of MiniMax-M3.1-Flash; this article's configuration follows the official recommendation of 524288 (512K), with 1M to be evaluated later when more code or document context needs to be retained.

• default_reasoning_summary set to none does not mean that thinking is disabled; supported_reasoning_levels still cannot contain none.

• The page does not specify the author, publication date, rate limits, timeouts, pricing, or subscription-region restrictions; these parameters cannot be added based on this configuration.

• This article does not contain a complete model prompt, so it is categorized as configuration and does not claim to provide a copyable prompt.

Reproduction notes

On 2026-10-21, the official page was opened through the Tabbit task M31 prompt05 codex; Option 2: Manual setup was expanded to extract config.toml, the model catalog, reasoning levels, context window, and restart instructions. The official MiniMax CLI documentation linked from the page was also reviewed to verify the wizard's backup behavior. Evidence was saved to /tmp/m31-codex-evidence.txt. Codex or MiniMax CLI was not installed locally; the wizard was not run, and ~/.codex or any other actual user configuration was not modified.