Mistral / Mistral Large 3 675B Instruct 2512
FP8 quantization for Mistral Large 3 675B Instruct 2512
The published profile is indicative; engineering measurements are completed for the selected model and hardware.
Before → After
Base releaseVerified checkpoint
675.000B
675.000B
QuantizationFP8
Quality retention target
Quality retention target
Production artifactArtifact
Lineage and benchmark report
Lineage and benchmark report
GPU targeting
Weights and KV cacheEstimated until measured during engineering
Estimated VRAMMeasured and certified results are delivered with the project.
Runtime packagingvllm