Gemma 3 12B

Training compute

8.6×10²³ FLOP

Parameters

12B

Published

Mar 12, 2025

Gemma 3 12B is an AI model developed by Google DeepMind (United States), first published in March 2025. It works in the language, vision and multimodal domain, on tasks such as language modeling/generation, question answering, translation and 4 more.

Training it took an estimated 8.6×10²³ FLOP of compute (estimation method: operation counting). The model has 12,000,000,000 parameters. It was trained on roughly 12T datapoints. Training ran on 6,144 Google TPU v4.

Access: Open weights (restricted use). Its weights are openly available. It is built on top of SigLIP 400M. Epoch AI rates the confidence of this record as confident.

Full record

Organization: Google DeepMind
Country of organization: United States
Domain: Language, Vision, Multimodal
Task: Language modeling/generation, Question answering, Translation, Chat, Quantitative reasoning, Visual question answering, Code generation
Training compute: 8.6×10²³ FLOP
Compute estimation method: Operation counting
Parameters: 12,000,000,000
Dataset size: 12T
Training hardware: Google TPU v4
Chips used: 6,144
Training power draw: 4.1 MW
Model accessibility: Open weights (restricted use)
Open weights: Yes
Base model: SigLIP 400M
Epoch confidence: Confident