DeepSeek Silently Launches V4 Pro Flagship Model That Nearly Matches Claude Fable 5 at 46 Times Lower Cost
Summary
DeepSeek silently launches its flagship V4 Pro model, updated to DeepSeek-V4-Pro-0813, nearly matching Claude Fable 5 on agent benchmarks at just $0.65 per million tokens — 46 times cheaper than Claude's $30 rate — with MIT-licensed weights publicly available on Hugging Face for independent verification.
Key Points
- DeepSeek quietly releases the finished version of its flagship V4 Pro model, updated to DeepSeek-V4-Pro-0813, with no formal announcement — only a version change on its API pricing page signals the upgrade.
- DeepSeek's own benchmark comparisons show Claude Fable 5 leads by just 5.3% on average across 10 agent benchmarks, but costs roughly 46 times more, with blended rates of $30 per million tokens versus $0.65 for V4 Pro.
- Chinese AI labs continue closing the performance gap with American frontier models at a fraction of the price, with DeepSeek's MIT-licensed weights publicly available on Hugging Face for independent verification — though no outside lab has benchmarked the new 0813 build yet.