Release2026-08-13
DeepSeek released the finished version of its V4-Pro model and, alongside it, an open-source toolkit called Harness that lets developers assemble their own coding agents as an alternative to Anthropic's Claude Code and OpenAI's Codex. From 16 August the company also replaces its single flat API price with higher peak and off-peak rates, so most workloads will cost more than they do today. Harness is a developer preview and its makers warn of changes that will break existing setups.
What changed
DeepSeek charged one flat API rate around the clock, cut 75% in late May.
What it unlocks
Building a coding agent on freely reusable software whose model, tools, sandbox and interface can each be replaced.
- V4-Pro output $0.87 to $1.98 off-peak
- peak output $3.96 per 1M tokens
- increases of 50% to 1,100%
- new pricing 16:00 UTC Aug 16
What you need to act on it
- developer skills to run the preview code
- an API key or self-hosted open weights for the models
Sources