SophyAI Super Local Intelligence

2026-10-07

Mistral Large 4: weights due on October 31

Mistral Large 4 is already in the catalog as an info card, with weights due on October 31, 2026. Meanwhile Mistral Small 4 runs on a single RTX PRO 6000.

MistralOpenWeightsvLLM

Mistral Large 4 is in the catalog as an info card: the weights are announced for October 31, 2026. Until they are out, it can't be started.

What it will need

The planned recipe uses vLLM on 8 B300 or B200 GPUs, with at least 512 GB of RAM and 2.5 TB of NVMe disk. It's a cloud-session model, not a Mac one.

As soon as the weights are public we verify them, benchmark them and update the recipe with real hourly cost and speed.

Meanwhile: Mistral Small 4

Mistral Small 4 can already be started:

Both run on vLLM and are published as a local OpenAI endpoint.

← All posts