TAU2
Measured May 14, 2026Source
Score
0.11
Models15 0 
Gemma 3 12B Instruct
GoogleGoogleUnited States
Gemma 3 12B Instruct is a 12-billion parameter, open-weight model from Google designed for efficient on-device and edge deployment. It features enhanced reasoning and instruction-following capabilities, along with native multimodal support for processing both text and images. The model offers a strong balance of performance, speed, and accessibility for developers.
ReasoningFastCheapMultimodal
Input / 1M tokens
$0.09Output / 1M tokens
$0.29Output tokens/s
24.14First-token seconds
1.86sSupported plans
0Benchmark history
Evaluations
Terminalbench Hard
Measured May 14, 2026Source
Score
0.01
Lcr
Measured May 14, 2026Source
Score
0.07
Ifbench
Measured May 14, 2026Source
Score
0.37
Aime 25
Measured May 14, 2026Source
Score
0.18
Aime
Measured May 14, 2026Source
Score
0.22
Math 500
Measured May 14, 2026Source
Score
0.85
Scicode
Measured May 14, 2026Source
Score
0.17
Livecodebench
Measured May 14, 2026Source
Score
0.14
Hle
Measured May 14, 2026Source
Score
0.05
Gpqa
Measured May 14, 2026Source
Score
0.35
Mmlu Pro
Measured May 14, 2026Source
Score
0.6
Artificial Analysis Math Index
Measured May 14, 2026Source
Score
18.3
Artificial Analysis Coding Index
Measured May 14, 2026Source
Score
6.3
Artificial Analysis Intelligence Index
Measured May 14, 2026Source
Score
8.8
Plan availability
Products and plans that support this model
No products or plans have been linked to this model yet.
Discussion
Thinking... Make sure you are connected to GitHub server