Mistral Drops 123-Billion-Parameter Weights for Local Use
Mistral AI released the full weights of its 123B model called Mistral Large 2 under an open license. The model matches or exceeds GPT-4 on MMLU, HumanEval, and GSM8K benchmarks while fitting on two 80 GB GPUs. Weights are downloadable today from Hugging Face.
Running frontier-class models locally removes API latency and data export concerns. Teams can now fine-tune or quantize the model for domain tasks without sharing proprietary documents. Cost calculations shift from per-token fees to electricity and hardware amortization.
Hugging Face hosts the weights and reports over 180 000 downloads in the first week. Independent developers have already produced 4-bit quantized versions that run at 35 tokens per second on a single RTX 4090.
Step 1: Visit huggingface.co/mistralai/Mistral-Large-2407 and accept the license terms. Step 2: Run pip install transformers torch and load the model with from_pretrained using device_map="auto". Step 3: Generate text with model.generate and observe output quality comparable to GPT-4 on your local GPU.