Mistral Small 3

Training compute

1.2×10²⁴ FLOP

Parameters

24B

Published

Jan 30, 2025

Mistral Small 3 is an AI model developed by Mistral AI (France), first published in January 2025. It works in the language domain, on tasks such as language modeling/generation, question answering, quantitative reasoning and 2 more.

Training it took an estimated 1.2×10²⁴ FLOP of compute (estimation method: operation counting). The model has 24,000,000,000 parameters. It was trained on roughly 8T datapoints.

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Mistral AI
Country of organization: France
Domain: Language
Task: Language modeling/generation, Question answering, Quantitative reasoning, Code generation, Translation
Training compute: 1.2×10²⁴ FLOP
Compute estimation method: Operation counting
Parameters: 24,000,000,000
Dataset size: 8T
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Confident