// AI news · 2026-10-29 · Hugging Face Blog
Mistral Releases Multimodal Mistral Large 4
Mistral releases Mistral Large 4, a multimodal MoE model with 1T total parameters and 1M token context.
Published 2026-10-29 08:00 (Beijing time)Source Hugging Face BlogAuto-summarised, fact-checked against source
01Summary
It excels in cybersecurity tasks like vulnerability reproduction and patching, achieving 82% on the Artificial Analysis Cyber Index. The model uses sparse attention and a Pixtral encoder, with strong performance in coding and agentic workflows. It is open-weight, enabling user control over deployment and refusal behavior.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- Mistral Large 4 supports text and image input with 1T total parameters and 1M context window.
- Scores 82% on Artificial Analysis Cyber Index, excelling in vulnerability reproduction and patching.
- Uses sparse attention with MoE architecture: 5 experts per token (4 selected + 1 shared).
- Open-weight design allows user control over deployment and harmful request refusal.
03Source
Welcome Le Chonk: Mistral Large 4 on Hugging Face
Hugging Face Blog · published 2026-10-29 08:00 Beijing time
Also reported by 3 other outlets
05Related news
- Mistral Releases 1T-Parameter Multimodal Model Le Chonk2026-10-29 · Mistral AI News
- Hugging Face Launches Speech & Decision Models for Apple2026-10-29 · Hugging Face Blog
- Hugging Face Launches Project ProtoStar2026-10-29 · Hugging Face Blog
- Procore uses OpenSharing for hourly data refreshes2026-10-29 · Databricks Blog
- Valve to Improve Steam Recommendation System2026-10-29 · IT之家
- Shanghai Proposes Ordered Deployment of Autonomous Driving in Taxi Services2026-10-29 · IT之家
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All AI news · Today