Unexpected <think></think> tokens at the beginning of generated code files when using MiniMax-M2.7

[caizhuojiang]

Hi NanoResearch team,

First of all, thank you for your excellent work on this project!

I encountered an issue while using MiniMax-M2.7 as the backend model. In the experiment stage, many generated code files contain unexpected <think></think> tokens at the very beginning, which causes the code to fail during execution.

Problem Description

• Generated code files start with <think></think>
• This leads to syntax errors and prevents the code from running properly

Configuration

Here is my current configuration:

{
  "research": {
    "base_url": "https://api.minimaxi.com/v1",
    "api_key": "xxx",
    "openalex_api_key": "xxx",
    "s2_api_key": "xxx",
    "template_format": "neurips2025",
    "execution_profile": "local_quick",
    "writing_mode": "hybrid",
    "max_retries": 2,
    "auto_create_env": true,
    "auto_download_resources": true,
    "ideation": { "model": "MiniMax-M2.7", "temperature": 0.5, "max_tokens": 16384, "timeout": 600.0 },
    "planning": { "model": "MiniMax-M2.7", "temperature": 0.2, "max_tokens": 16384, "timeout": 600.0 },
    "experiment": { "model": "MiniMax-M2.7", "temperature": 0.1, "max_tokens": 16384, "timeout": 600.0 },
    "code_gen": { "model": "MiniMax-M2.7", "temperature": 0.1, "max_tokens": 16384, "timeout": 600.0 },
    "writing": { "model": "MiniMax-M2.7", "temperature": 0.4, "max_tokens": 16384, "timeout": 600.0 },
    "figure_prompt": { "model": "MiniMax-M2.7", "temperature": 0.1, "max_tokens": 16384, "timeout": 300.0 },
    "figure_code": { "model": "MiniMax-M2.7", "temperature": 0.1, "max_tokens": 16384, "timeout": 300.0 },
    "evidence_extraction": { "model": "MiniMax-M2.7", "temperature": 0.1, "max_tokens": 16384, "timeout": 300.0 },
    "revision": { "model": "MiniMax-M2.7", "temperature": 0.3, "max_tokens": 16384, "timeout": 600.0 },
    "review": { "model": "MiniMax-M2.7", "temperature": 0.3, "max_tokens": 16384, "timeout": 300.0 },
    "figure_gen": {
      "model": "gemini-3.1-flash-image-preview",
      "image_backend": "gemini",
      "temperature": null,
      "timeout": 300.0
    }
  }
}

Question

Do you have any suggestions on how to prevent these <think></think> tokens from appearing in the generated code?

Any guidance would be greatly appreciated!

[YujunWu03]

Hi @caizhuojiang, thank you so much for your kind words and for trying out NanoResearch!
We've been able to reproduce this issue — reasoning models like MiniMax-M2.7 emit ... tokens before the actual response, which leak into generated code and cause syntax errors.We're currently testing a fix for this and will update you once it's verified. Thanks for reporting it!