Mistral ML4 goes open-weight: trained from scratch on just 10MW, claims top open model in code agents
_AndrewZhao · x · 2026-11-03
Mistral released ML4 (nicknamed Le Chonk) as an open-weight model with a technical report. VP of Science Pierre Stock says it was trained from scratch on only 10MW of compute (roughly 4,000 GB300s), a fraction of competitors' usage, all on Mistral's own hardware. After a year building their RL training machine, ML4 claims best-in-class open-weight performance on cyber defense, coding, and coding agents — and he publicly asks for 1 gigawatt of compute.
Related event: Mistral releases ML4 technical report: flagship trained on just 10 MW(7 posts)→
More from Infra
- B300/GB300 Blackwell financing deals smell like bogus multi-layer SPV schemes, warns insider — TheZachMueller · 2026-11-03
- Northwood raises $375M Series C as a16z touts space's early-internet moment — a16z · 2026-11-03
- Intel ships 14A PDK 0.9, a make-or-break milestone for its foundry ambitions — ryanshrout · 2026-11-03
- Compute growth driven entirely by R&D spend; customer-serving compute stays flat — teortaxesTex · 2026-11-03
- Strata inference engine silently caps thinking at 75% with an injected wrap-up line — Regular-Register-358 · 2026-11-03
- RTX 3090 tensor unlock on Windows finally boots: BitLocker fix, BF16 70→115 TFLOPS, 20% faster video gen — LavateLasManos666 · 2026-11-03