BEIJING — A new artificial intelligence model released by China's DeepSeek has ranked first among open-source models worldwide in web development coding performance.
DeepSeek's new model, V4.1-Flash-1017, released on the 17th, took the top spot among open-source models in web development coding evaluations on the AI benchmarking platform Arena, beating Moonshot AI's Kimi K3, Chinese media outlet Zhidongxi reported on the 20th. It also outperformed some U.S. closed-source models, including OpenAI's GPT-6 Astra, Anthropic's Claude Opus 5 and xAI's Grok 4.7. While top-tier U.S. closed-source models generally maintain an edge in overall performance, open models such as DeepSeek's have pulled ahead on certain coding tasks.
The bigger advantage is price. The new DeepSeek model costs $0.6 per million output tokens during off-peak hours, one-33rd the price of OpenAI's GPT-5.6 Sol at $20 and one-42nd that of Claude Opus 5 at $25. Peak hours run from 1 a.m. to 4 a.m. and 6 a.m. to 10 a.m. Coordinated Universal Time on weekdays, or 10 a.m. to 1 p.m. and 3 p.m. to 7 p.m. Korean time. Arena said the new model has entered the "Pareto frontier," meaning no other model delivers higher performance at a lower cost.
The new model is ultra-large, with 763 billion parameters, a figure that indicates the size of an AI model. Japanese outlet Giga IT reported that the model is too big to run on a typical personal computer, requiring DeepSeek's official application programming interface or an API from an outside provider. Ollama, a program that lets users run AI models on their own computers, is already offering an external API service for the model from U.S. servers.