Chinese language AI lab DeepSeek has introduced two preview variations of its latest massive language type, DeepSeek V4, a much-awaited replace to closing 12 months’s V3.2 type and the accompanying R1 reasoning type that took the AI international through hurricane.
The corporate says each DeepSeek V4 Flash and V4 Professional are mixture-of-experts fashions with context home windows of one million tokens each and every — sufficient to permit massive codebases or paperwork for use in activates. The combination-of-experts manner comes to activating just a sure choice of parameters according to job to decrease inference prices.
The Professional type has a complete of one.6 trillion parameters (49 billion energetic), which makes it the largest open-weight type to be had, outstripping Moonshot AI’s Kimi Ok 2.6 (1.1 trillion), MiniMax’s M1 (456 billion), and greater than double DeepSeek V3.2 (671 billion). The smaller, V4 Flash has 284 billion parameters (13 billion energetic).
DeepSeek says each fashions are extra environment friendly and performant than DeepSeek V3.2 because of architectural enhancements, and feature nearly “closed the distance” with present main fashions, each open and closed, on reasoning benchmarks.
The corporate claims its new V4-Professional-Max type outperforms its open-source friends throughout reasoning benchmarks, and outstrips OpenAI’s GPT-5.2 and Gemini 3.0 Professional on some duties. In coding festival benchmarks, DeepSeek mentioned each V4 fashions’ efficiency is “related to GPT-5.4.”
Then again, the fashions appear to fall relatively at the back of frontier fashions in wisdom exams, in particular OpenAI’s GPT-5.4 and Google’s newest Gemini 3.1 Professional. This lag suggests a “developmental trajectory that trails state of the art frontier fashions through roughly 3 to six months,” the lab wrote.
Each V4 Flash and V4 Professional enhance textual content best, not like a lot of its closed-source friends, which give enhance for working out and producing audio, video, and photographs.
Techcrunch tournament
San Francisco, CA
|
October 13-15, 2026
Significantly, DeepSeek V4 is a lot more inexpensive than any frontier type to be had nowadays. The smaller V4 Flash type prices $0.14 according to million enter tokens and $0.28 according to million output tokens, undercutting GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5. The bigger V4 Professional type, in the meantime, prices $0.145 according to million enter tokens and $3.48 according to million output tokens, additionally undercutting Gemini 3.1 Professional, GPT-5.5, Claude Opus 4.7, and GPT-5.4.
The release comes an afternoon after the U.S. accused China of stealing American AI labs’ IP on an commercial scale the use of 1000’s of proxy accounts. DeepSeek itself has been accused through Anthropic and OpenAI of “distilling,” necessarily copying, their AI fashions.
While you acquire via hyperlinks in our articles, we would possibly earn a small fee. This doesn’t have an effect on our editorial independence.
