DeepSeek Ships V4 Pro to Production — and Raises Prices Sharply
DeepSeek moves V4 Pro out of testing, open-sources its agent software — and hikes API prices steeply, introducing peak and off-peak rates.
Illustration · AI-generated (AI IN LIFE)
At a glance
- Input pricing rises to $0.66 (off-peak) / $1.32 (peak) per million tokens
- Cache-hit prices increase more than tenfold at peak
- New rates take effect August 16, 2026
- Terminal Bench 2.1: 72.1 → 87.9; DeepSWE: 12.8 → 62.7
- Agent software “Harness v0.1” now open source under MIT license
Chinese AI provider DeepSeek has moved its flagship model into production as V4-Pro-0813 — while announcing a substantial price increase. Starting August 16, time-based rates apply: V4 Pro input pricing rises from $0.435 to $0.66 per million tokens off-peak, and to $1.32 during peak hours.
Cache hits see the steepest increase: from $0.003625 to $0.022 off-peak and $0.044 at peak — more than tenfold at the high end. Peak windows align with Chinese business hours. DeepSeek attributes the move to heavily strained infrastructure and surging agent demand.
Technically, V4 Pro takes a clear step up: Terminal Bench 2.1 scores rise from 72.1 to 87.9, and DeepSWE jumps from 12.8 to 62.7. The context window stays at one million tokens; new additions include native support for OpenAI's Responses API and three reasoning-effort levels (“low,” “high,” “max”).
In parallel, DeepSeek open-sourced its agent software “Harness v0.1” under an MIT license. Its modular plugin system, called Cordis, makes tools, sandboxes, sessions, and even the UI swappable. More than 700 projects applied for the beta within three days.
The pricing offensive marks a strategy shift: DeepSeek entered the market as a price disruptor and is now monetizing premium performance — just as OpenAI cuts its prices. The ongoing fundraising ahead of a planned IPO forms part of the backdrop.
FAQ
Why is DeepSeek raising prices?
The company points to strained infrastructure from agent workloads and is prioritizing monetization ahead of a planned IPO rather than competing on price alone.
What exactly changes in the pricing?
From August 16, peak and off-peak rates apply: V4 Pro input costs $0.66 to $1.32 per million tokens, and cache hits rise to $0.022–$0.044.
What is DeepSeek Harness?
An open-source agent framework (MIT license) built on the Cordis plugin system, where every component — tools, sandboxes, sessions, UI — is swappable.
