NEWHow AI models draw the world map · Gecko Tests
Beta
Home/Models/Gemma 3 12B
Google DeepMind logo

Gemma 3 12B

by Google DeepMind · Released Mar 2025

Open SourceMultimodal
10.2
avg score
Rank #300
Compare
Better than 4% of all models
Context
131K tokens (~66 books)
Input $/1M
$0.05
Output $/1M
$0.15
Type
multimodal
License
Open Source
Benchmarks
6 tested
Data as of
About

Gemma 3 introduces multimodality, supporting vision-language input and text outputs. It handles context windows up to 128k tokens, understands over 140 languages, and offers improved math, reasoning, and chat capabilities,...

Tested on 6 benchmarks · BenchGecko score 10.2. Top scores: Chatbot Arena Elo — Overall (1341.8%), GPQA diamond (19.3%), OTIS Mock AIME 2024-2025 (16.6%).

Looking for similar performance at lower cost?
Llama 3.2 1B Instruct scores 11.7 (115% as good) at $0.03/1M input · 46% cheaper
Capabilities
math
16.6
#245 globally
knowledge
9.6
#286 globally
general
10.0
#191 globally
Benchmark Scores
Compare All
Tested on 6 benchmarks · Ranked across 4 categories
Score Distribution (all 312 models)
0255075100
▲ You are here
mathCompare math →
OTIS Mock AIME 2024-2025

Mock AIME (American Invitational Mathematics Exam) problems from OTIS. Tests mathematical competition performance.

16.6·
knowledgeCompare knowledge →
GPQA diamond

Graduate-level science questions written by PhD experts. Diamond subset contains questions where experts disagree, testing deep understanding.

19.3·
Chess Puzzles

Tactical chess puzzles testing pattern recognition and multi-move calculation. Measures strategic reasoning ability.

0.0·
arenaCompare arena →
Chatbot Arena Elo — Overall

Chatbot Arena overall Elo rating. Crowdsourced human preference ranking from blind head-to-head comparisons across all topics.

1342·
Quick compare:NVIDIA logovs Nemotron 3 Supereleutherai logovs Pythia 160mAlibaba logovs Qwen2 0.5B
Excellent (85+) Good (70-85) Average (50-70) Below (<50)
Similar Models
NVIDIA logo
Nemotron 3 Super
NVIDIA
10.2$0.08/1M
eleutherai logo
Pythia 160m
eleutherai
9.9n/a
Alibaba logo
Qwen2 0.5B
Alibaba
10.8n/a
Links
Info
Google DeepMindPricing explorerDevelopers · API
Research
Technical Report
Documentation
API DocsPlayground
Community
@Google DeepMind
Source Code
GitHubHugging Face
BenchGecko API
gemma-3-12b-it
Specifications
  • Typemultimodal
  • Context131K tokens (~66 books)
  • ReleasedMar 2025
  • LicenseOpen Source
  • StatusActive
  • Cost / Message~$0.000
Available On
Google DeepMind logoGoogle DeepMind$0.05
Categories
mathknowledgegeneral
Learn More
context-windowtransformeropen weightstokens
Share & Export
Tweet
Related Models
NVIDIA logoNemotron 3 Super
10.2
eleutherai logoPythia 160m
9.9
Alibaba logoQwen2 0.5B
10.8
SmolLM2 135M
9.0
Gemma 3 12B is an open-source multimodal AI model by Google DeepMind, released in March 2025. It has an average benchmark score of 10.2. Context window: 131K tokens.

Related Models

Nemotron 3 Super · NVIDIAPythia 160m · eleutheraiQwen2 0.5B · AlibabaSmolLM2 135M · Hugging Face TBGpt2 Medium · OpenAI

Benchmarks

Chatbot Arena Elo — OverallGPQA diamondOTIS Mock AIME 2024-2025DtbenchLmca

Related Pages

Google DeepMind · ProviderGoogle DeepMind · EconomyAll ModelsCompare ModelsPricingDevelopers · APIContext Window · Glossary

Key facts · as of 2026-10-05

  • Gemma 3 12B by Google DeepMind. BenchGecko score 10.2, rank 300 of 312 scored models (normalized average of public benchmark scores).
  • List price $0.0500 input · $0.15 output per 1M tokens (as of 2026-10-05).
  • Sold by 1 provider (as of 2026-10-05): DeepInfra (bf16) $0.0500 in / $0.15 out.

How to cite · data as of 2026-10-05

Gemma 3 12B · benchmarks, pricing and providers. BenchGecko, data as of 2026-10-05. https://benchgecko.ai/model/gemma-3-12b-it

Credit "Source: BenchGecko" with a link. Prices per provider and Gecko Tests are BenchGecko data (CC BY 4.0); benchmark scores keep their original source, listed in the JSON. JSON · llms.txt · MCP