China's AI power leads the world: Baitu Biotech xTrimo V3 Biological Model surpasses Evo2

The Evo2 biological AI model jointly released by Stanford University and Nvidia has sparked heated discussions. However, a detail in its paper highlights the rise of China's AI: Baitu Biotech's xTrimo series of models has been listed as a competitor with larger parameters but not yet open source.

Baitu Biotech, a company that is regarded as a pioneer in the basic model of life sciences by the US market, has continued to deepen its efforts in this field since 2020. In October 2024, the xTrimo V3 full-modal biological model launched, which set a new record for the world's largest basic life science AI model with 210 billion parameters.

The breakthrough of xTrimo V3 lies in its first implementation of biological data modeling from DNA, RNA, protein to cell levels, and supports 128K ultra-long DNA sequence analysis. More importantly, Baitu Biotech is about to open source its 100 billion parameter version, and its parameter scale will surpass Evo2 and become the world's largest open source biological model.

xTrimo V3: Double Leadership in Parameter Scale and Mode

The parameter volume of xTrimo V3 is more than five times that of Evo2. Its strong learning ability comes from the super-large-scale life science exclusive data map constructed by Baitu Biotechnology since 2020. This map integrates multi-dimensional and multi-modal biological information such as protein interaction networks, single-cell sequencing, genomics, and clinical data, and the data scale is more than a hundred times that of the industry benchmark.

Unlike overseas models such as ESM and Evo, which only focus on the single mode of protein or DNA sequences, xTrimo V3 can comprehensively model various biological data from molecules, metabolic networks to cells and even multicellular levels, and achieve cross-modal alignment technology to deeply analyze the underlying laws of life systems. At present, xTrimo V3 covers seven mainstream modes, achieving full-scale modeling from base pairs to cell clusters.

Innovative algorithm architecture to unleash the potential of super-large-scale data

Baitu Biotech's original MoE architecture and biological knowledge-guided training framework effectively avoid the problem that super-large-scale models are prone to "remember training errors" and maintain efficient learning ability while expanding parameters. In response to the differences in sequence lengths of different genes, xTrimo V3 pioneered the introduction of multi-window-scale attention mechanism and native double-stranded DNA modeling technology, showing stronger learning ability under the same conditions, and surpassing Evo1/Evo2 in core tasks such as gene mutation scanning.

Commercialization leads the new era of AI application in life sciences

Baitu Biotech has commercialized the xTrimo platform, and its technical system has generated significant value in the fields of AI target discovery, protein design, strain transformation, etc., and has reached a strategic cooperation with Sanofi, receiving a prepayment of US$10 million, with a total transaction amount of more than US$1 billion. At present, Baitu Biotech has served more than 400 users around the world, including top pharmaceutical companies, scientific research institutions and biological manufacturing companies, and its applications cover multiple fields such as drug research and development, agricultural and chemical industry, and environmental protection.

xTrimo has made breakthrough progress in the fields of antibody and cell gene therapy drug design, target discovery, microbial research, etc., showing huge application value and commercial potential.

The xTrimo V3 of Baitu Biotechnology marks an important milestone in the field of life science AI. The upcoming 100 billion parameter version will further promote global life science research and are expected to benefit more fields and bring broader prospects for human health and well-being.

The above is the detailed content of The hidden competition of the biological version of DeepSeek, the Chinese model is regarded as a stronger opponent, and Sanofi is betting on $1 billion!. For more information, please follow other related articles on the PHP Chinese website!