Mistral Large 4

	Mistral Large 4(mistral.ai)
	2028 points by Philpax 7 days ago \| 1210 comments
	tl;dr: Mistral is previewing Large 4 ("ML4"), a 1T-parameter (52B active) natively multimodal MoE model trained on 3,800 Grace Blackwell GPUs in European datacenters, with open weights promised by month's end. It claims competitive performance with top open-source models and leads in cybersecurity (notably because closed models like Claude Opus 5.5 and GPT-6 refuse such tasks), visual grounding, and legal/financial workloads. The release is backed by Mistral's €3B Series D and positioned as a sovereign, self-deployable alternative to US and Chinese frontier models.
	HN Discussion: ↑Benchmark testing confirms the model shows meaningful generational improvement in specific tasks like data analytics and visual grounding ↓Mistral's progress is mediocre and lags behind American and Chinese labs despite the article's optimistic framing ↑Even if not frontier-leading, the model matters for European sovereignty and provides valuable competition ↑The model is a solid catch-up effort with notable strengths in cybersecurity niches where closed models refuse •Questions about resource efficiency and what massive US investments are achieving if 4k GPUs can nearly match frontier results