Deepseek releases V4-Pro update and open-source agent software
Deepseek has moved its flagship V4-Pro model out of testing with build V4-Pro-0813. The model name, parameter count and one-million-token context window remain unchanged, and the company says existing integrations require no changes. The release adds native support for the OpenAI Responses API with Codex integration and offers low, high and max reasoning settings. Deepseek reports higher scores on Terminal Bench 2.1 and DeepSWE, while Artificial Analysis places V4-Pro at 53 on its Intelligence Index, below Claude Opus 5 at 63. The new weights have not yet been published.
The company also released Deepseek Harness v0.1 as a Developer Preview under the MIT license. Built on the Cordis plugin system, it supports replaceable components, persistent logs, and runs that can be resumed, branched or replayed. The software is launched through npx and a local web interface. The project is led by Cui Tianyi, who joined Deepseek from Jane Street in March 2026.
New API rates take effect on August 16 at 4:00 p.m. UTC. Off-peak V4-Pro input will cost $0.66 per million tokens and output $1.98; peak prices will be $1.32 and $3.96. Cache hits will cost $0.022 off-peak and $0.044 at peak.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.