Chinese AI lab DeepSeek has released the general availability version of its DeepSeek-V4-Pro model, build 0813. The update moves the system out of preview status, where it had been since April, and makes it broadly accessible to developers.
DeepSeek states that its model performs within 5.3% of Anthropic’s Claude Fable 5 on average across nine agent benchmarks. The company highlights a significant price difference, noting that its API costs are substantially lower than those of its US-based competitor.
Key facts
- DeepSeek-V4-Pro-0813 launched as a general availability release on August 12, 2026, without a formal announcement.
- The company claims its model won two specific internal tests and trails Claude Fable 5 by only 5.3% on average across nine benchmarks.
- DeepSeek charges $0.435 per million input tokens and $0.87 per million output tokens for the Pro model.
- In comparison, Anthropic lists prices of $10 per million input tokens and $50 per million output tokens for Claude Fable 5.
- Third-party experts have not yet independently verified the performance of the 0813 build, as DeepSeek’s data relies on internal infrastructure.
Why it matters
This release intensifies competition in the large language model market by offering a high-performance alternative at a fraction of the cost of leading US models. If independent testing confirms DeepSeek’s performance claims, developers and enterprises may find a viable, cost-effective option for deploying AI agents, potentially shifting market dynamics based on price-to-performance ratios.