DeepSeek has quietly replaced deepseek-v4-pro with its "0813" official release, marking the formal launch of a model that had been running in preview form since April. According to Decrypt, DeepSeek is presenting the update with a simple message: performance sits only about 5% behind Anthropic’s Claude Fable 5, while pricing is just a tiny fraction of that model’s cost.
DeepSeek says the gap is about 5.3% across nine agent benchmarks
Based on benchmark figures released by DeepSeek, V4 Pro comes with a 1 million-token context window and trails Claude Fable 5 by an average of about 5.3% across nine agent benchmarks. In two of those tests, DeepSeek said its model outperformed Claude Fable 5.
The report also notes that the "5% behind, 46 times cheaper" comparison comes from DeepSeek’s own published materials. Actual performance still awaits third-party verification.
Pricing is where the difference widens sharply
DeepSeek lists pricing at about $0.43 per million input tokens and $0.87 per million output tokens. Claude Fable 5, by comparison, is priced at $10 per million input tokens and $50 per million output tokens. Using those figures, the article says DeepSeek’s cost works out to roughly 1/46 of Claude Fable 5.
Official build posts clear gains over the preview version
Compared with the earlier preview release, the official version showed notable improvement on several hard metrics. Terminal Bench 2.1 rose from 72.1 to 87.9. DeepSWE jumped from 12.8 to 62.7. DSBench-FullStack also climbed from 41.8 to 71.1. The changes point to stronger coding and agent capabilities in the formal release.
ABMedia said this places DeepSeek in the middle of the latest price-driven contest around frontier models. The article also referenced a Chain News report published the same day on xAI’s Grok 4.6, which was described as offering an upgraded model at the same price while targeting frontier-model performance.
Pressure on top closed-source model pricing may keep building
The report said open-source and lower-cost competitors are continuing to close the performance gap with leading closed-source models. If that trend continues, pricing pressure on OpenAI and Anthropic is likely to intensify. For readers looking for more on DeepSeek’s background and controversies, the original article pointed to a separate Chain News guide.