Mistral Unveils Le Chonk: Large 4, a Multimodal Model with 1.05 Trillion Parameters
Mistral introduced its new flagship model Large 4, internally nicknamed Le Chonk. The multimodal MoE model has 1.05 trillion parameters, 49 billion active parameters, and supports context up to 1 million tokens.
Mistral Unveils Le Chonk: Large 4, a Multimodal Model with 1.05 Trillion Parameters
Mistral has introduced its new flagship model, Large 4. Internally, the company reportedly calls it Le Chonk, a reference to the Le Chaton Fat meme, a fake giant Mistral model that appeared on social media months ago.
Key details
- Architecture: MoE
- Parameters: 1.05 trillion total, 49 billion active
- Modalities: multimodal, including image analysis
- Context: up to 1 million tokens
- Availability: through the API in
public preview
According to Mistral's own tests, Large 4 is a new open SOTA model for the US and Europe. The company highlights coding, cybersecurity, and finance, and says the model outperforms some closed frontier models on image object detection tasks.
The model was trained from scratch for about two months on 4,000 Nvidia Grace Blackwell units. For a model of this size, that is a relatively compact cluster.
Weights are promised to be released on October 27.
Why it matters
AnalysisThe release strengthens competition in the open-weight AI market and gives developers a long-context multimodal model for coding, security, and finance workloads.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Google Gemini Pro (Model with Infinite Context)
The flagship working model from Google DeepMind featuring a record context window of over 2 million tokens. Capable of analyzing entire books, lengthy videos, and massive codebases in a single query.
LLM (Large Language Model)
A fundamental class of neural network architectures based on autoregressive transformers, predicting the probabilistic distribution of subsequent tokens and demonstrating emergent properties of abstract reasoning, code synthesis, and logical inference.
Part II. AI Agent Architecture: From Model to Memory and Tools
A deep engineering dive: tokenization, KV-cache, sampling determinism, the training stack, quantization, context engineering, the safety stack, production RAG, ReAct, and multi-agent swarm systems.
Alibaba launches Qwen4-Max with 1M-token context
Alibaba’s new flagship model is positioned for coding, game development and agent tasks, with a 1 million token context window and lower pricing than top US competitors.