Gemma 3 12B: benchmark results
Google's open-weight Gemma 3 12B, a mid-size multimodal model (March 2025). Provider: Google. Released 2025-03-12. Access: Open.
Aggregate Rating 1532 ± 5, rank #672 of 1656 rated models, from 1041 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Audio2Tool (Text Oracle) - Conversation | 29.9 | Exact match (%): share of queries whose predicted tool or or | 100 |
| Audio2Tool (Text Oracle) - Implicit Reasoning | 14.5 | Exact match (%): share of queries whose predicted tool or or | 100 |
| Audio2Tool (Text Oracle) - Needle-in-a-Haystack | 45 | Exact match (%): share of queries whose predicted tool or or | 100 |
| Audio2Tool (Text Oracle) - Parametric Commands | 12.3 | Exact match (%): share of queries whose predicted tool or or | 100 |
| BanglaMedVQA (CoT) | 35.75 | Exact-match accuracy (%; mean of the Chest X-Ray and MedICaT | 100 |
| BanglaMedVQA (CoT) - MedICaT | 40.5 | Exact-match accuracy (%; 1,000 questions on MedICaT figures; | 100 |
| BanglaMedVQA - MedICaT | 39.5 | Exact-match accuracy (%; 1,000 questions on MedICaT figures; | 100 |
| GeoDrive-Bench (Reasoning) | 58.5 | Overall accuracy (%) on 5,053 human-validated multiple-choic | 100 |
| HyperGVL - Vertex Counting | 99.8 | Accuracy (%) on counting the vertices of a hypergraph, 7,000 | 100 |
| TSCG 20 Tools | 95 | Scenario D, 20 tools, TSCG compilation: tool-selection accur | 100 |
| TSCG 50 Tools | 98 | Scenario D, 50 tools, TSCG compilation: tool-selection accur | 100 |
| OpenVLM MMT-Bench - Traffic Anomaly Detection | 75 | Score (%) | 99.5 |
Interactive version: theaggregate.ai/model?slug=gemma-3-12b · How It Works · Data refreshed daily, snapshot 2026-10-21.