Julien Chaumond praises Mistral's intelligence-per-GPU efficiency versus rival models

Julien Chaumond highlighted French, a post emphasizing Mistral's efficiency. The quoted post claims Mistral 4 reaches 72% of the intelligence of rival models using 3.8% of the GPUs, citing 3,800 Grace Blackwell GPUs against 100,000+ for GPT-6 Astra.