Mistral Large 4(昵称「Le Chonk」)是一款原生多模态模型,总参数约 1 万亿,每次处理请求时约有 490 亿参数处于激活状态。
该模型采用混合专家(MoE)架构。据报道,它在 Mistral 位于欧洲的数据中心从头训练而成。
目前模型可通过 API 预览;Mistral 计划于 10 月底发布可下载权重,发布前还将由网络安全专家和政府机构测试限制较少的版本。
What is Mistral AI’s Mistral Large 4 (“Le Chonk”), how was its 1 trillion-parameter mixture-of-experts model trained and designed, what canIllustrative AI-generated artwork; not a technical diagram of Mistral Large 4.
AI 提示
Create a landscape editorial hero image for this Studio Global article: What is Mistral AI’s Mistral Large 4 (“Le Chonk”), how was its 1 trillion-parameter mixture-of-experts model trained and designed, what can. Article summary: Mistral Large 4, nicknamed “Le Chonk,” is Mistral AI’s new flagship: a roughly 1-trillion-parameter, natively multimodal model available in an API preview since October 6, 2026. Mistral plans to release its weights later. Topic tags: general, general web, news, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts wi
openai.com
Mistral Large 4,昵称「Le Chonk」,是 Mistral AI 最新的旗舰模型:总参数约 1 万亿,但每次处理请求时激活的参数约为 490 亿。公司已于 2026 年 10 月 6 日开放 API 预览,并表示计划在本月晚些时候发布可下载权重。至于它是否整体胜过顶尖模型,目前仍应视为有待检验的性能说法,而不是定论。12732
它的架构与训练方式
Mistral 称 Large 4 是一款原生多模态模型,也就是从设计上支持处理文字以外的信息;公司特别强调其多模态理解能力,包括视觉任务。它也将模型定位于编程和智能体工作流等应用场景。110
Large 4 使用混合专家(MoE)架构。简单来说,并非每次请求都会调用模型的全部参数:虽然总参数规模约为 1 万亿,每次处理时约有 490 亿参数处于激活状态。因此,「万亿参数」指的是模型整体规模,并不代表每个请求都会用到全部参数。14
Mistral Large 4(昵称「Le Chonk」)是一款原生多模态模型,总参数约 1 万亿,每次处理请求时约有 490 亿参数处于激活状态。
该模型采用混合专家(MoE)架构。据报道,它在 Mistral 位于欧洲的数据中心从头训练而成。
目前模型可通过 API 预览;Mistral 计划于 10 月底发布可下载权重,发布前还将由网络安全专家和政府机构测试限制较少的版本。
What is Mistral AI’s Mistral Large 4 (“Le Chonk”), how was its 1 trillion-parameter mixture-of-experts model trained and designed, what canIllustrative AI-generated artwork; not a technical diagram of Mistral Large 4.
AI 提示
Create a landscape editorial hero image for this Studio Global article: What is Mistral AI’s Mistral Large 4 (“Le Chonk”), how was its 1 trillion-parameter mixture-of-experts model trained and designed, what can. Article summary: Mistral Large 4, nicknamed “Le Chonk,” is Mistral AI’s new flagship: a roughly 1-trillion-parameter, natively multimodal model available in an API preview since October 6, 2026. Mistral plans to release its weights later. Topic tags: general, general web, news, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts wi
openai.com
Mistral Large 4,昵称「Le Chonk」,是 Mistral AI 最新的旗舰模型:总参数约 1 万亿,但每次处理请求时激活的参数约为 490 亿。公司已于 2026 年 10 月 6 日开放 API 预览,并表示计划在本月晚些时候发布可下载权重。至于它是否整体胜过顶尖模型,目前仍应视为有待检验的性能说法,而不是定论。12732
它的架构与训练方式
Mistral 称 Large 4 是一款原生多模态模型,也就是从设计上支持处理文字以外的信息;公司特别强调其多模态理解能力,包括视觉任务。它也将模型定位于编程和智能体工作流等应用场景。110
Large 4 使用混合专家(MoE)架构。简单来说,并非每次请求都会调用模型的全部参数:虽然总参数规模约为 1 万亿,每次处理时约有 490 亿参数处于激活状态。因此,「万亿参数」指的是模型整体规模,并不代表每个请求都会用到全部参数。14