MiniMax-M3.1-Flash Claude Code Manual Setup and Model Verification

One-sentence takeaway

Option 2 on the official Claude Code integration page writes the Anthropic-compatible endpoint, Subscription Key, 512K automatic compaction window, and the MiniMax-M3.1-Flash[1m] model alias into ~/.claude/settings.json, then uses /status and /model for verification; this model always has thinking enabled and cannot turn it off.

Use cases

• Suitable tasks: Using MiniMax M Plan's MiniMax-M3.1-Flash in Claude Code for terminal programming and agent work.

• Unsuitable tasks: Scenarios that require thinking to be disabled or need to retain Anthropic environment variables that would overwrite the configuration file.

• Supported model version: MiniMax-M3.1-Flash; the 1M-context model name in the configuration is MiniMax-M3.1-Flash[1m].

• Applicable client, agent, or API: Claude Code; the MiniMax Anthropic-compatible endpoint https://api.minimax.io/anthropic.

• Recommended reasoning tier and parameters: Thinking is always enabled. The page does not specify an adjustable effort or other reasoning tier; CLAUDE_CODE_AUTO_COMPACT_WINDOW uses 524288.

Ready-to-use content

Clear high-priority environment variables before configuration

The official instructions require clearing the following variables before configuration; this article records the commands only and did not run them locally:

unset ANTHROPIC_AUTH_TOKEN
unset ANTHROPIC_BASE_URL

If ~/.bashrc or ~/.zshrc exports either variable, remove the corresponding line as well; otherwise a new shell will load them again. The official documentation says these two variables take precedence over the settings file and will overwrite the MiniMax configuration.

Option 2: ~/.claude/settings.json

Edit or create ~/.claude/settings.json, replacing <MINIMAX_API_KEY> with the MiniMax Subscription Key. The following is the copyable configuration provided on the official page:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.minimax.io/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "<MINIMAX_API_KEY>",
    "CLAUDE_CODE_AUTO_COMPACT_WINDOW": "524288",
    "ANTHROPIC_MODEL": "MiniMax-M3.1-Flash[1m]",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "MiniMax-M3.1-Flash[1m]",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "MiniMax-M3.1-Flash[1m]",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "MiniMax-M3.1-Flash[1m]"
  }
}

Testing or workflow steps

1. Confirm that Claude Code has been installed according to its official documentation; this article did not perform the installation.

2. Clear external exports of ANTHROPIC_AUTH_TOKEN and ANTHROPIC_BASE_URL, and check whether ~/.bashrc or ~/.zshrc will export them again.

3. Edit ~/.claude/settings.json according to Option 2, entering the Subscription Key and the model configuration above; this article did not write to the local file.

4. Enter the working directory, run claude, and select Trust This Folder.

5. Run the following in the Claude Code TUI:

   /status
   /model

6. Verify that /status shows ANTHROPIC_BASE_URL pointing to api.minimax.io/anthropic and that /model shows the current model as MiniMax-M3.1-Flash.

Original evidence and data

• The page title is Claude Code, and it says the goal is to connect MiniMax M-series models to Claude Code.

• Option 2 uses ~/.claude/settings.json as the settings file path. The credential field is ANTHROPIC_AUTH_TOKEN, its placeholder is <MINIMAX_API_KEY>, and the page calls it a Subscription Key.

• The official configuration sets ANTHROPIC_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL, ANTHROPIC_DEFAULT_OPUS_MODEL, and ANTHROPIC_DEFAULT_HAIKU_MODEL all to MiniMax-M3.1-Flash[1m].

• The value of CLAUDE_CODE_AUTO_COMPACT_WINDOW is the string "524288"; the page calls this the recommended 512K automatic compaction window.

• The verification target for /status is for ANTHROPIC_BASE_URL to point to api.minimax.io/anthropic; the verification target for /model is MiniMax-M3.1-Flash.

• The page explicitly states that MiniMax-M3.1-Flash always has thinking enabled and cannot turn it off.

• The page does not specify the author, publication date, effort / reasoning_effort, quotas, pricing, timeouts, or other parameter limits.

Scope and limitations

• This is a manual M Plan configuration for Claude Code; it cannot support inferences about configuration fields for other clients, MiniMax models, or the standard API.

• External ANTHROPIC_AUTH_TOKEN and ANTHROPIC_BASE_URL environment variables overwrite the settings file. If shell startup files still export them, the verification result may not represent the file configuration.

• MiniMax-M3.1-Flash[1m] is the Claude Code model alias given on the page; the /model verification text displays MiniMax-M3.1-Flash. Preserve the two forms according to their respective configuration fields.

• The page only says that thinking cannot be disabled. It provides no thinking budget, reasoning tier, or adjustable parameters, so undisclosed configuration cannot be added.

• This article does not contain a complete model prompt. It is therefore classified as configuration and does not claim that a copyable prompt exists.

Reproduction notes

On 2026-10-21, the official page was opened and read through the Tabbit exclusive task M31 prompt04 claudecode; Option 2 was selected manually, and its configuration code, environment-variable precedence, and verification steps were extracted. The evidence was saved to /tmp/m31-claudecode-evidence.txt. No installation command was run, and the local Claude Code configuration was not modified.