Mistral's roughly 1-trillion-parameter multimodal mixture-of-experts model with open weights, trained in Europe.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Mistral Large 4 (nicknamed "Le Chonk" or "Lonk") is Mistral's flagship model. YouTube reviewers describe it as a mixture-of-experts model of about 1 trillion total parameters, with roughly 49–52B active per token (reports differ). It is natively multimodal, supports both instruct and reasoning modes, and was reportedly trained from scratch in Europe on about 3,800 Nvidia Grace Blackwell GPUs. These details come from creator commentary, not from the provided official pages.
Reviewers report a 500k-token context window and API pricing of $1.36 input and $4.18 output per million tokens. They also report strong cybersecurity results and a high ranking among open models on some coding benchmarks. Overall intelligence is described as roughly on par with Chinese open models such as Kimi K3, GLM 5.3 and DeepSeek V4. One reviewer ranked it 25th on the Artificial Analysis index, and another found its price and intelligence uncompetitive. One tester reported friction with thinking tokens and context limits when using it in OpenCode.
Mistral says it positions its open-weight models for sovereign deployment. They can be run on-prem, in a customer VPC, or hosted by Mistral with in-region inference and dedicated GPU capacity. They can be customized with Forge and used through Studio and the Vibe apps. Creators said the weights were promised by the end of the month, so availability for download should be checked. The model is too large for most local use.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
1 trillion parameter MoE with 52B active, weights promised at end of month. Creator found its intelligence and price uncompetitive versus leading open models.
A 1-trillion-parameter open-weight model, roughly on par with Chinese open models like Kimi K3, GLM 5.3 and DeepSeek V4. Too large for local use.
Trillion-parameter open-weight mixture-of-experts model (49B active), natively multimodal, hybrid instruct/reasoning, trained from scratch in Europe on 3,800 Nvidia Grace Blackwell GPUs. Priced at $1.36/$4.18 per million tokens, strong on cyber benchmarks and second among open models on some coding benchmarks, but 25th on the Artificial Analysis index with a 500k context window. Creator tested it in OpenCode on a Rubik's cube simulation and had friction with thinking tokens and context limits; weights promised by end of month.
Closest neighbours by meaning, not just by category.
Model · NVIDIA
NVIDIA's open-weight ~550B mixture-of-experts LLM with a hybrid Mamba-transformer design, aimed at agentic workloads.
Model · Meituan
Meituan's open-weights 1.6T-parameter mixture-of-experts LLM (48B active), aimed at agentic coding and reasoning tasks.
Model · xAI
xAI's Grok 4.7 language model for coding and knowledge work, available via API, Cursor and Grok Build.
Model · DeepSeek
DeepSeek's V4 Pro large language model, available via API and as an open model, with low per-token pricing.
Model · Google
Google DeepMind's Gemini 4 Carbon, a frontier Gemini 4 language model with a staged rollout to paid users and API.
Model · Google
Smaller, faster, lower-cost Gemini 4 model from Google DeepMind with a 1M-token context window.
