DeepSeek ships V4-Pro — and raises API prices sharply
New rates take effect today: DeepSeek hikes API prices by up to roughly 1,100 percent, adds peak pricing, and open-sources its agent framework.
In short
Chinese AI provider DeepSeek has released the final build of its V4-Pro model (build 0813) while announcing a massive price increase: as of August 16, API rates rise by up to roughly 1,100 percent — the largest jump in the history of a company long known as the industry's price breaker.
At a glance
- API price increases of up to roughly 1,100%, effective August 16, 2026
- New peak/off-peak pricing aligned with Chinese business hours
- V4-Pro (build 0813): Terminal-bench 2.1 up from 72.1 to 87.9 points
- Artificial Analysis Index: 53 points — level with GLM-5.2, behind Opus 5 (63)
- Agent framework “Harness v0.1” open-sourced under the MIT license
Chinese AI provider DeepSeek has released the final build of its V4-Pro model (build 0813) while announcing a massive price increase: as of August 16, API rates rise by up to roughly 1,100 percent — the largest jump in the history of a company long known as the industry's price breaker.
At the core is a new peak/off-peak model aligned with Chinese business hours. V4-Pro input pricing rises from 0.435 to 0.66 dollars per million tokens off-peak and doubles to 1.32 dollars at peak times. Cache hits are hit hardest: from 0.003625 to as much as 0.044 dollars — roughly a factor of twelve, weighing especially on agents that perform many repeated lookups.
The model itself improves markedly: on Terminal-bench 2.1, V4-Pro climbs from 72.1 to 87.9 points, and on the DeepSWE benchmark from 12.8 to 62.7. On the Artificial Analysis Intelligence Index it rises from 45 to 53, drawing level with GLM-5.2 — while still trailing Western frontier models such as Claude Opus 5 (63 points).
In parallel, DeepSeek is opening its agent ecosystem: the “DeepSeek Harness v0.1” framework ships as a developer preview under the MIT license. It logs prompts, tool calls, and results, and lets developers resume, branch, and replay agent runs.
The pricing strategy marks a turning point: even Chinese providers can no longer — or no longer want to — subsidize the price war indefinitely. For companies built on DeepSeek, cost planning starts over today.
FAQ
Why is DeepSeek raising prices?
DeepSeek is introducing peak/off-peak rates to manage load spikes and to move its heavily subsidized API pricing closer to real compute costs.
Who is hit hardest?
Agent workloads with many repeated lookups: the price of cache hits rises by roughly a factor of twelve.
What is DeepSeek Harness?
An open-source agent framework (MIT license) that logs prompts, tool calls, and results, making agent runs reproducible.
