Chinese language startup says DeepSeek-V4-Professional beats all rival open fashions for maths and coding.
Printed On 24 Apr 2026
China’s DeepSeek has unveiled the newest variations of its signature synthetic intelligence-powered chatbot, a 12 months after its flagship mannequin despatched shockwaves by the worldwide tech scene.
The Chinese language startup launched preview variations of DeepSeek-V4-Professional and DeepSeek-V4-Flash on Friday because it touted its means to go toe-to-toe with US rivals equivalent to OpenAI and Google.
Beneficial Tales
record of 4 gadgetsfinish of record
Like DeepSeek’s earlier chatbots, V4-Professional and V4-Flash comply with an open-source mannequin, that means builders are free to make use of and modify the supply code at will.
DeepSeek-V4-Professional beats all rival open fashions for maths and coding, and trails solely Google’s Gemini 3.1-Professional, a closed mannequin, for world data, DeepSeek mentioned in an announcement on social media.
The “professional” model’s efficiency falls solely “marginally brief” of OpenAI’s GPT‑5.4 and Gemini 3.1-Professional, “suggesting a developmental trajectory that trails state-of-the-art frontier fashions by roughly 3 to six months,” the Hangzhou-based startup mentioned.
The “flash” mannequin has related reasoning skills to the “professional” model, whereas providing sooner response occasions and “extremely cost-effective” utilization pricing, the agency mentioned.
The discharge comes after DeepSeek-R1 shocked the tech sector upon its launch in January final 12 months with capabilities broadly comparable with these of ChatGPT and Gemini.
Marc Andreessen, a outstanding Silicon Valley enterprise capitalist with shut ties to United States President Donald Trump, hailed the mannequin’s launch on the time as “AI’s Sputnik second”.
The efficiency of the Chinese language-developed mannequin attracted specific consideration as its builders claimed to have spent lower than $6m on computing prices – a fraction of the multibillion-dollar budgets which might be ordinary in Silicon Valley.
Some tech analysts challenged DeepSeek’s account of working with such scant assets, arguing that the startup almost definitely had entry to better funding and extra superior chips than acknowledged.
DeepSeek’s arrival on the scene prompted blowback in some nations amid considerations about knowledge safety and Chinese language authorities censorship.
A number of US states, Australia, Taiwan, South Korea, Denmark and Italy launched bans or different restrictions on DeepSeek-R1 shortly after its launch, citing privateness and nationwide safety considerations.
