TikTok proprietor ByteDance, DeepSeek lead Chinese language push in AI reasoning

By Liam Mo and Brenda Goh

BEIJING (Reuters) – TikTok proprietor ByteDance on Wednesday launched an replace to its flagship AI mannequin geared toward difficult Microsoft-backed OpenAI’s newest reasoning mannequin merchandise, as a world race intensified to create AI fashions able to tackling advanced issues.

The corporate launched Doubao-1.5-pro, an improve to its flagship AI mannequin, which it claims outperforms OpenAI’s o1 in AIME, a benchmark take a look at that measures how effectively AI fashions perceive and reply to advanced directions.

ByteDance’s launch comes after Chinese language AI startup DeepSeek rolled out an open-source reasoning mannequin referred to as DeepSeek-R1 on Monday that it mentioned rivalled OpenAI’s o1 on a number of efficiency benchmarks.

DeepSeek drew widespread consideration in international AI circles final month after exams confirmed its V3 massive language mannequin outperformed these of OpenAI and Meta (NASDAQ:META), regardless of a smaller improvement finances and plans to cost customers quite a bit much less.

The developments in AI reasoning by ByteDance, DeepSeek and others is more likely to problem the market share of OpenAI and different massive language fashions when it comes to each efficiency metrics and costs charged to customers.

Different Chinese language corporations which have unveiled their very own reasoning fashions prior to now weeks embody Moonshot AI, Minimax and iFlyTek.

OpenAI triggered the race in AI improvement after it launched ChatGPT in November 2022 and its “Strawberry” sequence of AI reasoning fashions in September final yr. The latter are able to reasoning by way of advanced duties and fixing more difficult issues than earlier fashions in science, coding and math.

Final week, OpenAI CEO Sam Altman mentioned that they had finalized a model of its new reasoning AI mannequin o3 mini, and could be launching it in a few weeks.

DeepSeek proposed a cut-price payment providing for accessing and utilizing DeepSeek-R1, at 16 yuan ($2.20) per million tokens, significantly lower than OpenAI’s o1 438 yuan for a similar utilization.

ByteDance’s pricing is much more aggressive. Doubao-1.5-pro-32k prices 2 yuan per million tokens for output, whereas its extra highly effective Doubao-1.5-pro-256k model is priced at 9 yuan, in response to ByteDance’s cloud platform Volcano Engine.

($1 = 7.2798 Chinese language yuan renminbi)