독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · Mistral L4模型간단한 요점1 분 소요DisconfirmAI 편집부

Mistral L4 Preview Enters Agent Arena

Arena reports Mistral Large 4 preview scores -6.6%, ranking #43, ahead of predecessor.

2026-10-09 11:53 GMT+8

Mistral Large 4 preview recorded a -6.6% net improvement score in Agent Arena, ranking #43 overall.

The result is based on over 5,000 real-world agentic sessions. It ranks 11 positions higher than the previous variant, Mistral Medium 3.5 (-12.60%). At this score, it would rank #13 among open models.

Mistral AI announced the model on Oct 6 as a 1T-parameter, natively multimodal system, claiming it is the best open-weights model from US or Europe on aggregated benchmarks. Open weights are expected at end of October.

This data comes from Arena's official account for a preview model, not an independent third-party reproduction. Final open-weight performance may differ.

출처

  • x.com↗
주제

Mistral L4模型

法国AI实验室Mistral于2026年10月发布其旗舰模型Mistral Large 4(代号Le Chonk)的公开预览,并于10月29日正式开放权重。该模型拥有约1万亿总参数和520亿激活参数,采用混合专家(MoE)架构,支持多模态输…

4 단계4 건업데이트됨 2026-11-04 15:08

현재까지의 과정

  1. 2026-10-06Mistral发布Mistral Large 4公开预览版,宣称其为1万亿参数多模态模型,计划月底开放权重。
로그인 / 회원가입
독서리서치레이더투자 프레임워크
mistral.ai ↗
  • 2026-10-07The Deep View报道指出,Mistral将ML4定位为对抗供应商锁定的工具,特别强调其在网络安全防御中的自主可控优势。thedeepview.com ↗
  • 2026-10-09Arena公布ML4预览版在Agent Arena榜单中的表现,较前代提升11个排名,成为唯一进入前15名的欧洲实验室。x.com ↗
  • 아직 해결되지 않은 문제

    • Mistral Large 4在实际企业级网络安全事件响应中的长期稳定性与误报率如何?
    • 随着更多中国开源模型迭代,Mistral声称的‘西方最强’地位是否面临快速挑战?
    • Mistral Ecosystem Open-Weight License对Model-as-a-Service业务的限制具体如何影响初创公司采用?

    이 스레드의 다른 내용

    • French AI Firm Mistral Releases Open-Weight Model Mistral Large 4: Full Weights, Architecture Details and Benchmarks Available2026-10-29
    • Mistral Launches 1T Open Model2026-10-07
    전체 주제 읽기 →
    리서치 계속하기 →