Model Overview
DeepSeek-R1-0528-Qwen3-8B is a text generation and reasoning model with approximately 8 billion parameters. DeepSeek trained it by distilling Qwen3-8B Base using chain-of-thought data from DeepSeek-R1-0528. It is not the full DeepSeek-R1-0528 model. (api-docs.deepseek.com)
Capabilities and Uses
The model is suitable for mathematical reasoning, programming, and text generation, and can also be used for reasoning model research and small-model application development. Officially reported scores include 86.0% on AIME 2024, 76.3% on AIME 2025, 61.1% on GPQA Diamond, and 60.5% on LiveCodeBench. The AIME 2024 score is 10 percentage points higher than that of Qwen3-8B in the report; these results were obtained under specific test conditions and do not guarantee performance on real-world tasks. (huggingface.co)
The model weights are available in the official repository, which lists an MIT license, and can be used for local deployment; usage must comply with the license terms. (huggingface.co)
Limitations
- Outputs may still contain factual or reasoning errors; important results should be verified by a human.
- The model has the same architecture as Qwen3-8B but uses the DeepSeek-R1-0528 tokenizer configuration. The official instructions require using all configuration files from this model's repository when deploying it, rather than directly reusing the original Qwen3 configuration. (huggingface.co)
- The API capabilities, context length, or benchmark scores of the full DeepSeek-R1-0528 should not be directly treated as specifications of this 8B distilled model.
Release Date
May 29, 2025: Both the initial commit and file upload records in the official model repository show this date. DeepSeek's R1 update announcement is dated May 28, 2025, and also introduces this distilled model. (huggingface.co)
Pricing
As of November 18, 2026, DeepSeek's official domestic Chinese yuan (CNY) and international US dollar (USD) pricing pages had been checked, but no official API input or output prices for this model were verified on either page. Prices listed for other models on the two pages cannot be used to infer this model's price; self-deployment may also incur costs for computing resources and the like. (api-docs.deepseek.com)
Official Sources
- Model card and weights repository
- Model repository commit history
- DeepSeek update announcement
- Official domestic pricing page; checked on: November 18, 2026
- Official international pricing page; checked on: November 18, 2026