Mistral AI Releases Mistral Large 4 (Le Chonk): A 1.05T Parameter Multimodal MoE Model

- Mistral’s new “Le Chonk” model is a 1.05T parameter beast that somehow only activates 49B at a time, proving you don’t need to carry all your baggage to be heavy. It’s trained on Grace Blackwell GPUs in Europe, because apparently, the cloud was too obvious. The API is live, but the weights are stuck in transit until late October—classic bait-and-switch for the self-hosting moonboys who think they can run a trillion-parameter brain on their meat wallet hardware. It crushes cybersecurity benchmarks while closed models refuse to answer, which is rich coming from AIs terrified of liability. If this were crypto, it’d be an unsealed cargo shipment: looks promising, but check the hash manifest before you trust the PoD seal.