Mistral Large 4 goes open weights: 1.05T params, tops non-China open models on AA Index
ArtificialAnlys · x · 2026-10-29
Mistral has open-sourced its flagship Mistral Large 4 (1.05T total / 52B active params), scoring 26 on the Artificial Analysis Intelligence Index v5 — narrowly the #1 open weights model developed outside China, though 6 points behind DeepSeek V4.1 Flash 1017 and GLM-5.3, mainly on coding/agentic tasks (15% on FrontierCode 2.0 vs 29%/22%).
Key details:
- Token efficiency: 82K output tokens per task, 39% fewer than Preview, on the open-weights Pareto frontier
- Hallucinations down: 28% on AA-Omniscience v1.1 (from 41%), though accuracy remains low at 26%
- #2 on Coding Agent Index: 48 score using 8.5M tokens/task, about half of GLM-5.3's usage
- 50% price cut: $0.70/$2.10 per 1M input/output tokens; cost per task down 63% to $2.06
- 1M context window, text+image input
More from Models
- Investor bets speech-to-speech will be massive, praises Phonic's model — graceisford · 2026-10-30
- User Praises OpenAI for Doubling Pro Usage and Including API in Pro Sub — Angaisb_ · 2026-10-30
- OpenAI doubles usage limits for Pro tiers and Business Premium, effective immediately — Polymarket · 2026-10-30
- Gemini 4 Argon tops 4 robotics tasks on RLE-Bench, ranks #2 behind GPT-6 Astra — daibond_alpha · 2026-10-30
- Rumor: GPT-6.5 Bel comes from a 10-trillion-parameter pretraining run — VraserX · 2026-10-30
- Tracking chart logs 25 straight days of OpenAI shipping, day by day — burny_tech · 2026-10-30