")[1].strip()

")[1].strip()

## Summary

DeepSeek-R1 provides one of the most permissive licensing structures in the open-weight LLM ecosystem:

- The **base DeepSeek-R1 model** uses the **MIT License**, permitting unrestricted commercial and research use with minimal attribution requirements.
- **Qwen-derived distillates** (1.5B–32B) fall under **Apache 2.0**, requiring license preservation and change documentation.
- **Llama-derived distillates** (8B, 70B) carry **Meta's Llama 3.1/3.3 licenses**, which include specific acceptable use restrictions.
- Proper **runtime configuration** (temperature 0.6, no system prompts, step-by-step cues) is essential for activating the model's reasoning capabilities.

## Frequently Asked Questions

### Can I use DeepSeek-R1 commercially without paying royalties?

Yes. The MIT license explicitly permits commercial use, modification, and distribution without royalty fees. You can embed DeepSeek-R1 in products, offer it as a service, or fine-tune it for specific business applications. The only requirement is preserving the copyright notice and permission notice as specified in `/README.md#L56-L57`.

### Do the distilled smaller models have different licensing terms than the main R1 model?

Yes. While the base DeepSeek-R1 uses MIT, the distilled variants inherit licenses from their foundation models. The Qwen-derived models (1.5B, 7B, 14B, 32B) use Apache 2.0, and the Llama-derived models (8B, 70B) use Meta's Llama 3.1 and 3.3 licenses respectively. Always verify the specific license for the variant you deploy, as documented in `/README.md#L58-L62`.

### What attribution is required when redistributing DeepSeek-R1 or its derivatives?

You must include the original MIT license text and copyright notice when redistributing the base model. For Qwen-derived distillates, Apache 2.0 requires you to include the license text, state significant changes, and retain all notices. For Llama-derived models, follow Meta's attribution requirements. Failure to include these notices violates the license terms and could result in termination of your usage rights.

### Why does the documentation recommend against using system prompts with DeepSeek-R1?

The model was specifically trained to perform reasoning within the user prompt context, and prepending a system prompt can interfere with the chain-of-thought activation patterns. According to `/README.md#L91-L93`, placing all instructions directly in the user prompt ensures the model maintains its step-by-step reasoning capability and produces coherent outputs. This architectural constraint differs from many other LLMs that benefit from explicit system-level instruction separation.