On January 22, Beijing (Reuters) - ByteDance, the owner of TikTok, unveiled an updated version of its leading AI model, Doubao-1.5-pro, with the aim of rivaling Microsoft-backed OpenAI's latest reasoning model offerings in a competitive global race to develop AI models capable of solving intricate problems.
ByteDance claims that Doubao-1.5-pro surpasses OpenAI's o1 in AIME, a benchmark test evaluating AI models' comprehension and response to complex instructions.
This move by ByteDance follows Chinese AI startup DeepSeek's launch of an open-source reasoning model, DeepSeek-R1, designed to compete with OpenAI's o1 on various performance benchmarks.
Recent tests showed that DeepSeek's V3 large language model surpassed those of OpenAI and Meta, despite operating on a smaller budget and planning to offer services at more affordable prices.
The advancements in AI reasoning by ByteDance, DeepSeek, and other companies are poised to challenge OpenAI and other major language models in terms of performance and pricing.
Notable Chinese firms that have introduced their own reasoning models in recent weeks include Moonshot AI, Minimax, and iFlyTek.
OpenAI set the pace in AI development with the launch of ChatGPT in November 2022 and its "Strawberry" series of AI reasoning models in September last year, capable of tackling complex tasks in science, coding, and math.
OpenAI's CEO, Sam Altman, announced the upcoming launch of its new reasoning AI model, o3 mini, in a couple of weeks.
DeepSeek proposed a competitive fee of 16 yuan ($2.20) per million tokens for accessing and utilizing DeepSeek-R1, significantly lower than OpenAI's charge of 438 yuan for o1 under similar use.
ByteDance's pricing is even more aggressive with Doubao-1.5-pro-32k priced at 2 yuan per million tokens for output, and Doubao-1.5-pro-256k at 9 yuan, according to ByteDance's cloud platform, Volcano Engine.
(1 USD = 7.2798 CNY)