mistralai

Devstral 2 123B

Mistral's 125B-parameter coding-agent model with a 256K-token context window.Modified MITHugging Face

No server in stock right now has the 143 GB this model needs for inference.

License restrictions

Not licensed for companies with over USD 20M consolidated monthly revenue without a commercial agreement from Mistral.

From model cards and vendor documentation. Lower precision needs less memory, with slightly lower quality. Estimates are marked.

What you want to do	Precision	Minimum	Basis
Run the model Arithmetic: 119.4 GiB weights (mostly FP8 with some BF16, one of the repo's two identical weight formats) x 1.2 runtime overhead. No GB figure published; card recommends vLLM server deployment.	FP8, 8-bit	143.3 GB	Estimated