Key Points

• Deepseek has released an updated version of its V4-Pro model. It scores higher on agent benchmarks but still trails top models like Claude Opus 5 in overall rankings.
• The company is also releasing "Deepseek Harness," open-source software that turns language models into autonomous agents through a modular plugin system.
• API prices are going up, with new time-based rates that make usage outside Chinese business hours cheaper. Repeated data retrievals are getting pricier.

Deepseek has moved its flagship product out of the testing phase, released its proprietary agent software as open source, and announced higher API prices at the same time.

The deepseek-v4-pro endpoint now delivers build V4-Pro-0813. The model name, parameter count, and one-million-token context window remain unchanged, and Deepseek says existing integrations will keep running without any tweaks. In the app and on the web, the model is available under "Expert Mode." A new addition is native support for the OpenAI Responses API with Codex integration. Reasoning effort can be set to three levels: "low," "high," and "max," with Deepseek recommending the middle setting for everyday agent use.

According to Deepseek's own comparison table, Terminal Bench 2.1 scores jumped from 72.1 to 87.9, and DeepSWE scores went from 12.8 to 62.7. On several agent benchmarks, the model beat Claude Opus 4.8.

Artificial Analysis backs up the improvement but also puts it in context. V4-Pro climbs from 45 to 53 on the Intelligence Index, tying GLM-5.2. That's still behind Muse Spark  at 57, Qwen 3.8 Max at 58, and Kimi K3 at 60. Claude Opus 5 sits at the top with 63 points. Deepseek hasn't published the weights for the new build yet, and the April preview version is still up on Hugging Face.

The update was also a response to the smaller V4 Flash model closing in on the flagship. At the end of July, Deepseek shipped update 0731 for V4 Flash, which practically matched the Pro Preview on the Artificial Analysis Intelligence Index while costing a fraction of the price.

Agent harness ships under the MIT license

Alongside the model update, Deepseek Harness v0.1 is shipping as a Developer Preview under the MIT license. The open-source agent software is pitched as an alternative to OpenAI's Codex and Claude. It's built on the newly released Cordis plugin system, where all features are swappable plugins, from tools and sandboxes to sessions and the UI. A continuous session log tracks every prompt, every tool call, and every result. Runs can be resumed, branched, and replayed.

Minimal mode strips things down to the shell and file editor, and Deepseek uses this setup for its own benchmark runs. The software launches via npx through a local web interface, though Deepseek warns of compatibility issues. The project is led by Cui Tianyi, who joined Deepseek from quantitative trading firm Jane Street in March 2026. When the team put out a call for beta testers in early August, 712 projects signed up within three days.

API prices are going up, especially for cache hits

The new rates kick in on August 16 at 4:00 p.m. UTC. Deepseek announced the switch to peak and off-peak pricing at the end of June but didn't share specific figures or a date at the time. Time-based rates have been around since February 2025, when the company offered a discount on V3 and R1 during nighttime hours.

Off-peak usage costs half as much. Peak hours run from 1 a.m. to 4 a.m. and 6 a.m. to 10 a.m. UTC, lining up with the Chinese workday. For users in Europe, nearly the entire afternoon falls under the lower rate.

During off-peak hours, V4-Pro input goes from $0.435 to $0.66 per million tokens, and output jumps from $0.87 to $1.98. During peak hours, those rates double to $1.32 and $3.96. Cache hits are seeing the steepest increase, going from $0.003625 to $0.022 off-peak and $0.044 at peak. That shrinks the cache discount from about one-hundred-twentieth to one-thirtieth of the regular input price. For agents that repeatedly read the same files, this is the most expensive part of the change.

The new pricing partially undoes the price cut Deepseek rolled out in May, and cache hits will actually cost more than they did before that reduction. The price hike comes as the company is raising new capital and preparing for an initial public offering.

X/Deepseek

Harness