DeepSeek has released the V4-Pro-0813 model, with API pricing set at 0.025 CNY per million tokens for cached input, 3 CNY for uncached input, and 6 CNY for output, with a concurrency limit of 500. Compared to the Flash version, it is more expensive but has a lower concurrency limit, and prices may rise soon.
SynthePulse Insight · AI deep readingMembers
Version 1 · 2 sources
On August 13, DeepSeek officially open-sourced the V4-Pro-0813 model and launched the Harness code agent framework. This article, based on multiple sources, outlines its pricing, technical specifications, open-source strategy, and industry impact.
On August 12, X user Xiangyang Qiaomu posted that DeepSeek-V4-Pro-0813 had appeared in the official API documentation, joking that the naming date did not match the actual date. Subsequently, on August 13, DeepSeek officially open-sourced the model and simultaneously launched the Harness framework.
Some users interpreted the '0813' in the name as possibly referencing a historical event, but this is only speculation; the official naming reason has not been provided.
This article's information primarily comes from X user screenshots and reports from Geek Park citing IT Home. Pricing and concurrency data are source claims and have not been officially confirmed by DeepSeek. Technical specifications and benchmark results come from cited reports and are source claims. The open-source fact and Harness launch are confirmed, but details should be verified against official documentation.
Primary report
Primary source
Also covered by 1 sources