When do the new API prices take effect?

August 16 at 4:00 p.m. UTC. Peak hours (higher rates) run from 1 a.m. to 4 a.m. and 6 a.m. to 10 a.m. UTC; off-peak (lower) rates apply the rest of the day.

How much do cache hits cost under the new pricing?

Off-peak: $0.022 per million tokens; peak: $0.044 per million tokens. This is up from $0.003625 off-peak before the change, and costs more than they did before DeepSeek's May price cut.

What is the open-source agent software, and who can use it?

Deepseek Harness v0.1, released as a Developer Preview under the MIT license. It launches via npx through a local web interface and is pitched as an alternative to OpenAI's Codex and Claude. When DeepSeek called for beta testers in early August, 712 projects signed up within three days.

Large Language ModelsOpen-Source AIAI Business & IndustryTHE DECODERAug 13, 2026Aug 14, 2026, 04:01 JST

DeepSeek ships V4 Pro out of testing, open-sources agent software, raises API prices

Send on LINE

3 Key Points

What happened
DeepSeek moved its V4-Pro flagship model to production (build V4-Pro-0813), released its proprietary agent harness as open-source software under the MIT license, and announced higher API pricing effective August 16. On internal benchmarks, Terminal Bench 2.1 scores jumped from 72.1 to 87.9, and DeepSWE scores rose from 12.8 to 62.7.
Why it matters
The production release stabilizes the model for developers building agentic AI applications. Open-sourcing the agent harness positions DeepSeek as a competitor to OpenAI's Codex and Claude-based agent frameworks. However, the price increases—especially for cache hits, which now cost $0.022 per million tokens off-peak (up from $0.003625)—raise costs significantly for applications that repeatedly process the same data.
What to watch
Peak pricing runs 1–4 a.m. and 6–10 a.m. UTC (Chinese business hours), so European users see lower rates in the afternoon. The company is raising capital and preparing for an initial public offering, which may signal further pricing changes ahead.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

DeepSeek's move to production with V4-Pro signals confidence in the model's stability and performance gains. The Terminal Bench 2.1 jump from 72.1 to 87.9 and DeepSWE improvement from 12.8 to 62.7 address a competitive vulnerability: the smaller V4 Flash model, updated in late July, had nearly matched the Pro Preview on the Artificial Analysis Intelligence Index while costing far less. By releasing an improved Pro build, DeepSeek aims to preserve its flagship's margin over the budget tier.

The open-sourcing of Deepseek Harness under the MIT license is a strategic bet on developer adoption. Agents—AI systems that use tools and reasoning iteratively—are becoming central to enterprise AI applications, and OpenAI and Anthropic have dominated this space with Codex and Claude-based frameworks. DeepSeek's contribution, built on a "swappable plugins" architecture via the Cordis system, offers developers a modular alternative. The rapid uptake (712 beta signups in three days) suggests genuine interest, though the project is still in Developer Preview and carries compatibility warnings.

The pricing shift reveals the economics behind DeepSeek's aggressive expansion. Cache-hit fees are rising most steeply—from roughly one one-hundredth to one thirtieth of the input price—which will hurt use cases that read large static files repeatedly. This partly reverses DeepSeek's May price cut and signals that the company is optimizing for revenue as it prepares for capital-raising and an initial public offering. The adoption of time-based pricing aligned to Chinese business hours (1–4 a.m. and 6–10 a.m. UTC) reveals where core demand lies.

FAQ

When do the new API prices take effect?: August 16 at 4:00 p.m. UTC. Peak hours (higher rates) run from 1 a.m. to 4 a.m. and 6 a.m. to 10 a.m. UTC; off-peak (lower) rates apply the rest of the day.
How much do cache hits cost under the new pricing?: Off-peak: $0.022 per million tokens; peak: $0.044 per million tokens. This is up from $0.003625 off-peak before the change, and costs more than they did before DeepSeek's May price cut.
What is the open-source agent software, and who can use it?: Deepseek Harness v0.1, released as a Developer Preview under the MIT license. It launches via npx through a local web interface and is pitched as an alternative to OpenAI's Codex and Claude. When DeepSeek called for beta testers in early August, 712 projects signed up within three days.

THE DECODERRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.