DeepSeek releases improved V4-Pro model, raises API prices, and open-sources its agent software

DeepSeek has released an updated version of its V4-Pro model, designated V4-Pro-0813, now available via the deepseek-v4-pro endpoint. While model name, parameter count, and context window remain unchanged, the new build introduces native support for OpenAI's Responses API with Codex integration, and allows reasoning effort to be set to low, high, or max. In DeepSeek's own benchmarks, Terminal Bench 2.1 rose from 72.1 to 87.9 points and DeepSWE from 12.8 to 62.7, surpassing Claude Opus 4.8 on several agent tests. However, Artificial Analysis shows a continuing gap: V4-Pro's Intelligence Index rose from 45 to 53, tying with GLM-5.2, but still behind Muse Spark 1.2 (57), Qwen3.8 Max (58), Kimi K3 (60), and Claude Opus 5 (63). The weights of the new build have not been released. The update follows the V4 Flash model's July 31 update, which nearly matched the Pro preview at a fraction of the price. Additionally, DeepSeek released DeepSeek Harness v0.1 as a developer preview under the MIT license, built on the Cordis plugin system, offering an alternative to OpenAI Codex and Claude Code. The software features an extensible plugin architecture and a session log for resuming or forking runs. Finally, DeepSeek announced new API pricing effective August 16 at 16:00 UTC, with peak and off-peak rates. Off-peak, V4-Pro input costs $0.66 per million tokens (up from $0.435), output $1.98 (up from $0.87); peak rates are $1.32 and $3.96. Cache hit prices rise sharply from $0.003625 to $0.022 off-peak and $0.044 peak, significantly reducing the cache discount. This partially retracts May's price cuts and comes as DeepSeek seeks new funding and prepares for an IPO.