DeepSeek V4.1 Flash Cuts China US AI Model Gap to 3%, Bloomberg Intelligence Says
Summary
The gap between top Chinese and US AI models narrows to 3%, a record low, according to Bloomberg Intelligence analyst Robert Lea, driven by DeepSeek's September release of V4.1 Flash. The model scores 77.3 on LiveBench's agentic coding sub-benchmark versus 66.1 for Anthropic's Claude Fable 5.1 Max Effort. DeepSeek separately claims 74.2% on DeepSWE v1.1 without independent validation.
Key Points
- Bloomberg Intelligence senior analyst Robert Lea reports a record-low 3% gap between top Chinese and American AI models, driven by DeepSeek's September release of V4.1 Flash.
- DeepSeek V4.1 Flash scores 77.3 on LiveBench's agentic coding sub-benchmark versus 66.1 for Anthropic's Claude Fable 5.1 Max Effort on the October 4 snapshot.
- The gap had stood at roughly 9% in May 2026 and about 15% earlier in the year, and DeepSeek's own report claims 74.2% on DeepSWE v1.1 without independent validation.