Mistral 128B Benchmarked on 3x3090

·2026.05.04 09:46

Key point

Running Mistral Medium 3.5 128B Q3_K_M locally on 3x3090 (72GB VRAM) measured 9.67t/s.

1 / 2

Ran Mistral-Medium-3.5-128B-Q3_K_M locally on 3x3090 (72GB VRAM).

In the Python example, it output Mandelbrot generation code, with 504 tokens / 52s / 9.67 t/s shown at the bottom.
In the SVG example, it generated SVG code for a drawing of a pelican riding a bicycle, and in the HTML example, it generated code for an Atari 800XL fan page.
This is a real-world speed benchmark of running a 128B-class model on 3 consumer GPUs.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.