DeepSeek’s updated V4 Pro model narrows performance difference with Anthropic’s Claude 3.5 to 5% while maintaining a 4,500% price advantage.
DeepSeek released its final V4 Pro model, reducing the performance gap with Anthropic’s Claude 3.5 Fable to just 5% across nine agent benchmarks. The updated model, now in general availability, previously ran as a preview since April but remained untested in its final form by independent labs.
Claude 3.5 Fable costs $10 per million input tokens and $50 per million output, compared to DeepSeek V4 Pro’s $0.435 and $0.87. The pricing structure remains unchanged, but the underlying model weights have been updated. DeepSeek’s July announcement indicated the Pro API was still in preview, with the official release following quietly this week.
No independent benchmarking of the new 0813 build has been conducted yet, leaving the updated performance claims unverified outside the company’s own data.