DeepSeek-R1-0528: New Update Brings Top-Tier Reasoning, Logic, and Coding Skills
DeepSeek has just announced a powerful upgrade to its flagship AI model, DeepSeek R1, unveiling version DeepSeek-R1-0528. This new release represents a leap forward in reasoning depth, coding ability, and math-solving skills, pushing the model closer than ever to matching giants like OpenAI's O3 and Gemini 2.5 Pro.
- Mathematics: AIME 2024 accuracy reached 91.4%, HMMT 2025 scores nearly doubled
- Programming: LiveCodeBench performance jumped from 63.5% to 73.3%
- Logic & QA: GPQA-Diamond saw a 9.5% boost, "Humanity's Last Exam" accuracy more than doubled
- User Experience: Lower hallucination rates and smoother function calling
- Web Interface: Available at chat.deepseek.com with "DeepThink" mode
- Developer API: Accessible through platform.deepseek.com
- Local Deployment: Supports system prompts and simplified reasoning triggers
- Open Source: Released under MIT License for full commercial use and distillation
Sharper Brain, Deeper Thought
Thanks to enhanced post-training algorithms and more computing power, DeepSeek-R1-0528 has made major progress in handling complex reasoning tasks. For example, in the notoriously tough AIME 2025 math test, the model's accuracy jumped from 70% to 87.5%. It now uses nearly double the thinking power per question, averaging 23,000 tokens versus 12,000 previously, a sign of deeper and more nuanced reasoning.
Better in Code, Logic, and Real-World Reasoning
The model shines across a wide range of benchmarks:
- Mathematics: AIME 2024 accuracy rose to 91.4%, and HMMT 2025 scores nearly doubled.
- Programming: In LiveCodeBench, performance leaped from 63.5% to 73.3%.
- Logic & QA: GPQA-Diamond saw a 9.5% boost, and "Humanity's Last Exam" accuracy more than doubled.
Beyond performance, users will enjoy a lower hallucination rate, smoother function calling, and a more intuitive experience when using the model for coding support (also known as "vibe coding").
DeepSeek R1 Benchmark Evolution
Interactive comparison between original and improved versions
General
Code
Math
Tools
Qwen3-8B Learns to Think
The update also brings something for open-source fans: a distilled version called DeepSeek-R1-0528-Qwen3-8B. It transfers the thinking patterns of DeepSeek-R1 into the lightweight Qwen3 8B model, achieving state-of-the-art resultsamong models in its class. It even rivals much larger models like Qwen3-235B on tests like AIME 2024.
How to Try It
You can interact with DeepSeek-R1-0528 at chat.deepseek.com by activating the "DeepThink" mode, or access it via API at platform.deepseek.com. For developers, running it locally is also easier than ever. The latest version now supports system prompts and no longer requires special formatting to trigger advanced reasoning.
Open-Source and Ready to Use
All DeepSeek-R1 models, including this latest version, are released under the MIT License, allowing for full commercial use and distillation. This means startups, researchers, and developers can build on top of the model freely.
