Model Overview
DeepSeek-R1-Distill-Qwen-1.5B is an approximately 1.5-billion-parameter reasoning distillation model released by DeepSeek. It is based on Qwen2.5-Math-1.5B and fine-tuned using samples generated by DeepSeek-R1. The model weights have been made public, and the model card specifies the MIT license. (github.com)
Capabilities and Uses
The model is intended for text generation and reasoning. It can be used for mathematical problem-solving, programming assistance, and reasoning experiments; its weights can also be downloaded for local deployment and research. In official evaluations, this model achieved 28.9% pass@1 on AIME 2024 and 83.9% pass@1 on MATH-500. These results depend on the official evaluation settings and do not represent performance on all real-world tasks. (github.com)
Limitations
The model may generate incorrect answers or code, and important results require human verification. Its evaluation performance cannot be directly equated with that of the larger R1 or other distilled models. The officially provided configuration and tokenizer should be used for deployment; actual resource consumption depends on the deployment method. (github.com)
Release Date
2025-01-20: Official listing date. The commit history of DeepSeek's official model repository for this model shows an initial commit, file uploads, and release-related commits on that date. (huggingface.co)
Source note: In the current data, the independent initial release date of this specific model has not yet been independently verified. The date above is based on its official model repository's listing record and is not asserted to be an independently confirmed initial release date; DeepSeek's announcement on the same day confirmed the release of R1 and the simultaneous open-sourcing of six small distilled models. (huggingface.co)
Pricing
As of 2026-11-20, DeepSeek's official domestic RMB and international USD API pricing pages have been checked, but no official input or output token prices applicable to this model were verified, so no amounts are listed. Prices for the official R1 release and other models cannot be directly treated as prices for this distilled model; self-deployment may still incur computing resource costs. (api-docs.deepseek.com)
Official Sources
- DeepSeek-R1 official project and model information (github.com)
- Official model card for this model and commit history (huggingface.co)
- DeepSeek-R1 release announcement (api-docs.deepseek.com)
- Official pricing checks: domestic RMB page and international USD page, both checked on 2026-11-20. (api-docs.deepseek.com)