mistralai/Mistral-Large-4-1T-A52B

just landedmodel

Mistral Large 4 is our largest and most capable model to date: a 1 trillion-parameter natively multimodal Mixture-of-Experts model with 49 billion active parameters per token (52 billion including embeddings and output layers), trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's… Mixture-of-experts, 52B active (128 experts, 4 per token). Context: 1M tokens (config). License: other. Vendor-reported: "While a generalist model, it shows especially good performance on workloads such as cybersecurity, finance, and law, and it surpasses frontier closed models on visual grounding." Vendor-reported: "Vision: analyzes images, charts, technical drawings, and documents, with state-of-the-art visual grounding among open weights."

Vendor-reported: While a generalist model, it shows especially good performance on workloads such as cybersecurity, finance, and law, and it surpasses frontier closed models on visual grounding.
Vendor-reported: Vision: analyzes images, charts, technical drawings, and documents, with state-of-the-art visual grounding among open weights.

Org: mistralai
Released: 2026-09-30
Status: just landed
License: other (restricted)
Parameters: total, 52.0B active
Architecture: MoE · 128 experts, 4/token · MistralLarge4ForConditionalGeneration
Context: 1,048,576 tokens
Task: image-text-to-text · vision
Signal: 1278 HF likes · 1239 downloads

Runs on

real quant files; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.

Q4 ~606.5 GB: no listed device
Q8 ~1119.1 GB: no listed device
FP16 ~2104.8 GB: no listed device

Links

First published here 2026-09-30 · updated 2026-10-30 · markdown · JSON