Model positioning
DeepSeek-R1-Distill-Qwen-7B is a dense reasoning model with approximately 7 billion parameters released by DeepSeek. It is based on Qwen2.5-Math-7B and fine-tuned using samples generated by DeepSeek-R1. It is a standalone model in the R1 distillation series, not the full DeepSeek-R1. (github.com)
Capabilities and uses
The model is intended for text tasks such as mathematical, coding, and natural-language reasoning, and can be used for problem-solving, programming assistance, and reasoning research. In official evaluations, the model achieved a pass@1 of 55.5% on AIME 2024 and 92.8% on MATH-500; these results were obtained under specific evaluation conditions and do not guarantee performance on real-world tasks. (github.com)
The official model weights are available for local deployment and research. The official repository states that the code and weights are licensed under the MIT license and notes the original license of its Qwen base model. The relevant licenses should be consulted when using the model. (huggingface.co)
Limitations
The model may produce incorrect answers or repetitive or incoherent output; important results require human verification. The official recommendation is to set the sampling temperature between 0.5 and 0.7, with 0.6 recommended; deployment should use the officially provided configuration and tokenizer settings. (github.com)
Date
2025-01-20: Date the official model repository became available. DeepSeek's official release announcement declared on that date that six small distilled models were being open-sourced simultaneously; the official model repository's commit history for this specific model also records file uploads and release commits on that date. (api-docs.deepseek.com)
Source note: In the current data, whether this specific model had an earlier, separate initial release before it became available has not been independently verified. The date the official model repository became available, as noted above, is used here and is not presumed to be the date of an earlier initial release. (huggingface.co)
Pricing
As of 2026-11-20, after reviewing both the official DeepSeek Chinese pricing page and the official English pricing page, no official API input or output unit prices for this specific model could be verified, so neither RMB nor USD prices have been verified. Prices for other models listed on the pricing pages, as well as historical prices for the original DeepSeek-R1, cannot be applied to this 7B distilled model. Computing resource costs for self-deployment depend on the environment used. (api-docs.deepseek.com)