Skip to main content

​
Be Clear and Specific with Instructions

The M2.1 model responds well to clear and specific instructions. Clearly stating your expected output format, content, and style will help achieve more accurate results.
🪫 Less Effective
Create a visualization website
🚀 More Effective
Create an enterprise-grade data visualization website. Integrate as many rich analytical features and interactive functions as possible, going beyond basic display formats to build a fully-featured digital solution.

​
Explain Your Intent to Improve Performance

When giving instructions to M2.1, tell it “why.” When the model understands your purpose, it can provide more accurate answers.
🪫 Less Effective
Do not use document symbols
🚀 More Effective
Your response will be read aloud by a text-to-speech model, so present it in plain text format and avoid using document symbol formatting
M2.1 can generalize from examples. When you explain the context and reasoning clearly, it can follow your thought process and provide more relevant answers.

​
Focus on Examples and Details

Show it what you want with a standard “template” example; clearly point out what mistakes to avoid.
🪫 Less Effective
Write an engaging product description for a smart thermos.
🚀 More Effective
Please write a product description following this example:

[Good example: This desk lamp uses full-spectrum LED technology that simulates natural morning light to gently wake you up. It features 6 brightness levels to meet your different needs for reading, working, and resting.]

Please avoid vague descriptions like this:

[Bad example: This desk lamp is great, the light is comfortable, and the design is nice.]

Now, write a description for a 'smart thermos'.

​
Long Task Reasoning and State Tracking

The M2.1 model has excellent state tracking mechanisms. By focusing on limited goals each time rather than processing everything in parallel, it effectively maintains coherence and direction in long-sequence thinking.

​
Single-Window Context Awareness

M2.1 is equipped with context awareness features for efficient task execution and optimized context management.
When using tools that support context compression (such as Claude Code), it’s recommended to control the number of tokens in system prompts. M2.1 may terminate tasks early when approaching context capacity thresholds.

​
Multi-Window Workflow

1

Phased Processing

First window sets up the framework (writing, testing, creating scripts), second window iterates through to-do items
2

Structured Testing

Ask M2.1 to create tests.py or tests.json to track tests, helpful for long-term iteration
3

Initialization Scripts

Create init.sh to start servers and run tests, avoiding repetitive operations in new windows
4

Restart vs Compression

Use compression for single tasks, restart with a fresh window for multiple or new tasks
5

Maximize Context Usage

Prompt M2.1 to efficiently complete each part before continuing, making full use of tokens
Recommended System Prompt: 
This is a very lengthy task. It's recommended that you make full use of the complete output context to handle it—keep the total input and output tokens within 200k tokens. Make full use of the context window length to complete the task thoroughly and avoid exhausting tokens.