Mistral Large 4 launches: 1T multimodal MoE with 1M context, SGLang offers Day-0 support on Vera Rubin
NVIDIAAI · x · 2026-10-29
Mistral has released Mistral Large 4 (nicknamed Le Chonk), its largest model to date: a natively multimodal 1T-parameter MoE unifying instruct, reasoning, coding, and agentic workflows. It claims SOTA among open weights on security tasks, strong document understanding, and a 1M context window.
SGLang announced Day-0 support and is the first to serve it on NVIDIA Vera Rubin, with NVFP4 and DSpark speculative decoding supported from day one.
More from Infra
- 10 coding agents tested on local Qwen 3.8 27B: Pi wins on efficiency at 9.2k tokens per task — latent_lurker8587 · 2026-10-30
- Banks Shelve $5-7B Junk-Bond Sale for Anthropic-Leased Texas Data Center on Grid Audit — rohanpaul_ai · 2026-10-30
- Weight compression on the GPU: researcher proves his own instinct wrong — realDanFu · 2026-10-30
- Cached tokens explained: agent loops overpay for the same prefix, 50x savings — Arindam_1729 · 2026-10-30
- GB300 NVL72 runs DeepSeek Flash at ~1000 tok/s/user, ~$0.07 per 1M tokens TCO — bookwormengr · 2026-10-30
- Sam Altman: self-replicating AI-built data centers are 'within grasp' — Eliv_nurotic · 2026-10-30