Model Overview
DeepSeek-R1-Distill-Llama-8B is a distilled reasoning model with approximately 8 billion parameters released by DeepSeek. It is based on Llama 3.1 8B and fine-tuned using samples generated by DeepSeek-R1; it is not the full DeepSeek-R1 large model. (github.com)
Capabilities and Uses
The model is intended for text reasoning and can be used for mathematical problem-solving, code-related tasks, and reasoning research. In the official report, this model's AIME 2024 pass@1 is 50.4, MATH-500 pass@1 is 89.1, and LiveCodeBench pass@1 is 39.6; these are results under specific evaluation conditions and do not represent performance in all scenarios. Model weights are available for download, making it suitable for local deployment and research. (github.com)
Limitations and License
- The model may generate incorrect answers; users should verify important results. The official guidance warns that improper sampling settings may lead to repetitive or incoherent output and recommends setting the temperature to 0.5–0.7. (github.com)
- The official repository states that the code and model weights are under the MIT license, while noting that this model derives from a base model originally released under the Llama 3.1 license; the relevant license terms should be checked before use and redistribution. (github.com)
Release Date
2025-01-20: Date of the official release announcement. DeepSeek announced on that date that it was also making six small distilled models available; the official release commit on the same date explicitly listed DeepSeek-R1-Distill-Llama-8B and its download link. This date records the official release event and is not an end-of-month placeholder. (api-docs.deepseek.com)
Source note: In the current data, an earlier initial release date independent of the official release event described above has not been verified; the basis for 2025-01-20 is the official announcement and the release commit on the same date, and this does not establish that no earlier availability existed. (github.com)
Pricing
As of 2026-11-20, DeepSeek's official domestic RMB pricing page and international USD pricing page have been searched, but API input or output prices specific to this model have not been verified, so no amounts are listed. Prices for other models on the official pages do not apply to this model; self-deployment also requires consideration of the cost of the infrastructure used. (api-docs.deepseek.com)