Mistral Large 2 Ships 123 Billion Parameters Under an Open License
Mistral released a 123 billion parameter model under an Apache 2.0 license. The weights run on a single 80 GB H100 or A100, or on two RTX 4090 cards with 4-bit quantization. Developers avoid API fees and data exfiltration risks.
Open weights remove the pay-per-token barrier that previously limited experimentation. Teams can now benchmark internal data locally before committing to any vendor contract. The change forces product decisions to prioritize data control rather than model access.
Mistral AI published the weights at mistral.ai/news/mistral-large-2407. Early adopters report running the model at 35 tokens per second on a single A100 for customer support fine-tunes.
Step 1: Visit huggingface.co/mistralai/Mistral-Large-2407 and download the 4-bit GGUF file. Step 2: Install llama.cpp with CUDA support and run the command llama-server -m mistral-large-2407.Q4_K_M.gguf -c 8192. Step 3: Point your local client at http://localhost:8080 to obtain responses without external API calls.