Google's new math reasoning AI model
谷歌最强数学推理 AI 模型曝光：100 万词元上下文，专攻数学难题
This is the first detailed look at Google's new math-focused AI model, which could be a key competitor to existing models like DeepSeek's Math model.
Google is building an experimental AI model called Mathematica based on its DeepThink V3 model.
It is designed for complex math problems and will have a 1 million token context window.The model is currently in an unstable experimental state and is only available for internal testing.
谷歌最强数学推理 AI 模型曝光：100 万词元上下文，专攻数学难题
IT之家 9 月 18 日消息，消息源 @lyraxana 于 9 月 16 日在 X 平台发布推文，爆料称谷歌正基于 DeepThink V3 模型， 构建内部代号为 Mathematica 的实验 AI 模型，主要针对繁重计算、复杂的符号问题求解特别优化，预估将成为谷歌最强数学推理 AI 模型。 API 数据显示，该模型内部标识符为“models / deepthink-mathematica-tf-raw-thoughts ”，支持高达 100 万个词元（token）的上下文窗口，同时输出限制为 65,536 个词元。 IT之家注：上下文窗口指模型单次处理时可读取的文本单位总量；输出上限指模型单次回答可生成的文本单位数量。 泄露配置还标注“UNSTABLE_EXPERIMENTAL”。这一状态名称表明，该版本属于不稳定实验版本。模型同时带有“Teamfood”标签，是谷歌用于内部员工测试的术语，意味着相关服务处于内部测试范围。 TestingCatalog 指出，谷歌此前已推出面向高阶数学竞赛的 Gemini DeepThink IMO 模式，并在 9 月 15 日推出 Gemini 3.8 Live 和 Live Extended Thinking。
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Google, DeepThink, Mathematica, math reasoning.
Source attribution: curated from IT之家.
