DeepSeek's R1-0528 Model Rivals OpenAI and Google

DeepSeek has announced that its latest AI model, R1-0528, now matches the performance of leading models from OpenAI and Google. The Chinese AI startup highlighted that the upgraded model surpasses Alibaba's Qwen3 and is on par with OpenAI's o3 and Google's Gemini 2.5 Pro, .

The R1-0528 model features enhanced reasoning and creative writing capabilities, making it adept at crafting argumentative essays and fiction. Additionally, its coding capabilities have been improved, with a 50% reduction in "hallucinations," or instances of generating misleading information. The model's accuracy on complex reasoning tasks has increased to 87.5%, up from 70% in earlier versions.

DeepSeek continues to embrace an open-source approach, releasing the model weights to the public under a permissive MIT license. This strategy allows for commercial use and broader adoption, challenging the notion that only massive computational investments can yield top-tier AI performance. The company has also introduced a distilled variant of the R1-0528 model, capable of operating on a single GPU while outperforming other models in several benchmarks.