DeepSeek V4-Pro-0813 official version drops as a midnight bombshell: it can take on America's top-tier Fable model, priced at 3x Flash.

Tonight is another feast for AI enthusiasts: Musk released Grok 4.6, Qwen open-sourced Qwen 3.8 Max, and DeepSeek also dropped a nuclear bomb late at night, quietly launching the official version of V4 Pro—DeepSeek-V4-Pro-0813.

DeepSeek-V4-Pro-0813 is now officially live, with both the API and Chat updated. The calling method remains unchanged—use deepseek-v4-flash or deepseek-v4-pro to access the latest version.

What everyone cares about most is, of course, performance. The benchmark scores for V4-Pro-0813 have been released, comparing it not only against the preview version and the Flash-0731 version, but also against two top-tier large models, Opus 4.8 and Fable 5.

Let's get straight to the conclusion: V4-Pro-0813 has no problem surpassing Opus 4.8, and it trades blows with Fable 5 in performance. Whether it surpasses it or not, it's at least truly competitive now.

Here's an additional benchmark comparison including the K3 version: V4-Pro-0813 is roughly on par with the K3-to-Fable tier.

Finally, regarding pricing: V4-Pro-0813 remains priced at 3 times that of V4-Flash overall. The official team previously mentioned several times that they would adjust API pricing, but the current page still shows the old prices—no peak pricing, no mention of how much it will increase. Presumably, an official notice will come later.