Gemma 3 12B IT

Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions.

Native PDF input

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.27
Output: $0.27

Compare provider prices

Specifications

Context window: 128K
Max output: 131.1K
Parameters: 12B
Avg output (7d): 154 tokens; Longer than 11% of models

Benchmarks

Artificial Analysis

LMArena

Vectara

Sourced from Artificial Analysis.

Intelligence Index

3.8

Better than 0% of models compared

Coding Index

5.8

Better than 5% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

0.2%

Better than 1% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

0 Elo

Better than 3% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

-405 Elo

Better than 1% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

1.0%

Better than 7% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

8.3%

Better than 14% of models compared

Reasoning

HLE

Humanity's Last Exam

4.2%

Better than 16% of models compared

IFBench

Instruction-following benchmark

36.7%

Better than 30% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

16.4%

Better than 2% of models compared

LiveCodeBench

Contamination-free coding benchmark

13.7%

Better than 13% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

18.3%

Better than 21% of models compared

AIME

American Invitational Mathematics Examination

22.0%

Better than 48% of models compared

Math-500

Diverse mathematical problem solving benchmark

85.3%

Better than 54% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

59.5%

Better than 23% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

34.9%

Better than 12% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

0.8%

Better than 13% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

10.8%

Better than 8% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation