DeepSeek has officially moved its flagship V4 Pro model out of preview and into general availability, shipped an open-source coding agent called Harness, and announced sharp API price increases — all on the same day. The triple announcement signals the Chinese AI lab's transition from an aggressive disruptor to a maturing commercial player preparing for an initial public offering.
According to The Decoder, the new build, designated V4-Pro-0813, retains the same parameter count and one-million-token context window as the April preview version, but posts significant benchmark improvements. DeepSeek's own comparison tables show Terminal Bench 2.1 scores jumping from 72.1 to 87.9, while DeepSWE scores surged from 12.8 to 62.7. On several agent benchmarks, the model reportedly beat Claude Opus 4.8.
The independent Artificial Analysis Intelligence Index offers a more measured picture. V4 Pro climbed from 45 to 53 points on the index, tying GLM-5.2 but still trailing leaders like Kimi K3 at 60 and Claude Opus 5 at 63. Decrypt noted that Anthropic's Claude Fable 5 is roughly five percent better than V4 Pro — but at approximately 4,500 percent of the price.
Harness v0.1: An Open-Source Challenger to Claude Code
Alongside the model update, DeepSeek released Harness v0.1 as a Developer Preview under the MIT license. The open-source agent software is pitched as a direct alternative to OpenAI's Codex and Anthropic's Claude Code. It is built on DeepSeek's new Cordis plugin system, where all features — tools, sandboxes, sessions, and the UI — are swappable plugins.
The Decoder reported that Harness includes a continuous session log tracking every prompt, tool call, and result, with runs that can be resumed, branched, and replayed. A minimal mode strips the interface down to a shell and file editor, which DeepSeek uses for its own benchmark runs. The project launches via npx through a local web interface.
Harness is led by Cui Tianyi, who joined DeepSeek from quantitative trading firm Jane Street in March 2026. When the team called for beta testers in early August, 712 projects signed up within three days, The Decoder reported.
Cybersecurity Strength, Multimodal Gap
The South China Morning Post reported that V4 Pro's update underwhelmed in most categories — except cybersecurity, where the model showed notable strength. Tech Times noted that benchmark claims still await independent verification, with the model going to general availability before third-party labs have fully validated DeepSeek's self-reported scores.
The model also lacks multimodal capabilities, a gap acknowledged even by sympathetic reviewers. 36Kr noted that apart from multimodal features, V4 Pro met all expectations for a frontier model.
API Pricing: A Dramatic Reversal
The most consequential change for developers is the pricing overhaul. New rates take effect on August 16 at 4:00 p.m. UTC, implementing the peak and off-peak pricing structure DeepSeek first previewed in June.
During off-peak hours, V4 Pro input costs rise from $0.435 to $0.66 per million tokens, while output jumps from $0.87 to $1.98. During peak hours — aligned with the Chinese workday — those rates double to $1.32 and $3.96. Bloomberg and Fortune both reported that DeepSeek is increasing prices by multiples of previous rates.
The steepest increase hits cache hits, which jump from $0.003625 to $0.022 off-peak and $0.044 at peak. The Decoder noted that this shrinks the cache discount from roughly one-twentieth to one-thirtieth of the regular input price. For agent workflows 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. Reuters reported that the move comes as DeepSeek is raising new capital and preparing for an initial public offering.
Context: V4 Flash Put Pressure on the Flagship
The official release was partly a response to internal competition. At the end of July, DeepSeek shipped an update for its smaller V4 Flash model that practically matched the Pro preview on the Artificial Analysis Intelligence Index — at a fraction of the cost. The V4 Pro update re-establishes the flagship's lead, though by a narrower margin than before.
DeepSeek has also added native support for the OpenAI Responses API with Codex integration to V4 Pro. Reasoning effort can now be set to three levels: low, high, and max, with DeepSeek recommending the middle setting for everyday agent use.
The company has not yet published weights for the V4-Pro-0813 build. The April preview version remains available on Hugging Face.
What This Means for the AI Landscape
DeepSeek's simultaneous product launch, open-source release, and price hike represent a pivotal moment. The company that once shocked the industry with rock-bottom prices is now signaling it needs sustainable revenue. The Harness agent places DeepSeek in direct competition with Anthropic's Claude Code and OpenAI's Codex in the fast-growing coding agent market.
Meanwhile, the price increases could reshape the economics of AI development for companies that built cost-sensitive pipelines around DeepSeek's previously cheap API. With the IPO on the horizon, DeepSeek is making clear it intends to compete on quality and ecosystem — not just on price.
For the latest breaking AI news and in-depth coverage of the AI industry, AI Buzz Wire continues to track model releases, pricing changes, and open-source developments as they happen.
Stay Ahead of AIThe AI landscape moves fast. From DeepSeek's latest moves to frontier model benchmarks, our team covers the stories that matter. Read more AI news →
