The latest update from DeepSeek, a Chinese startup known for its open source reasoning AI model R1, has introduced DeepSeek-R1-0528. This significant update brings the free and open model closer in reasoning capabilities to proprietary paid models like OpenAI’s o3 and Google Gemini 2.5 Pro. The update aims to enhance performance on complex reasoning tasks in various fields and offers features for developers and researchers.
One of the key highlights of DeepSeek-R1-0528 is its improved performance on challenging reasoning tasks. The update has led to notable enhancements in benchmarks such as the AIME 2025 test, where accuracy increased from 70% to 87.5%. The model now handles deeper reasoning processes and has seen improvements in coding performance as well.
In addition to performance upgrades, DeepSeek-R1-0528 introduces new features to enhance the user experience. Support for JSON output and function calling aims to simplify integration for developers, while improvements in front-end capabilities promise a smoother interaction for users. The update also reduces the model’s hallucination rate and introduces system prompts for a more streamlined deployment process.
