Mistral AI Launches 1T-Parameter 'Le Chonk', Claims Open-Model Leadership

What happened: Paris-based Mistral AI has released its Mistral Large 4 model, nicknamed 'Le Chonk', a 1.05 trillion total parameter (49 billion active) multimodal mixture-of-experts (MoE) system. Trained on 3,800 NVIDIA Grace Blackwell GPUs, the model is now in public preview with a 1 million-token context window and promises open weights by late October. On benchmarks, 'Le Chonk' outperformed GPT-6 Astra on the Finance Agent v2 test (54.7 vs 53.5) but lagged behind Anthropic's Claude Opus 5.5 on agentic and coding tasks. API pricing (per Mistral docs) is $0.68/M input tokens and $2.09/M output, with Decrypt reporting higher, unverified prices.

Why it matters: Mistral positions 'Le Chonk' as the strongest open-weight model in the West, though it ranks #9 globally among open models per the Vals Index. Its open release and competitive pricing signal intensifying pressure on closed AI incumbents. However, performance leadership is nuanced—Mistral's model leads only on select benchmarks, and trails on others, highlighting the fragmented nature of current AI supremacy claims.

Source: Decrypt