Gemma 4 12B (IT): benchmark results
Google's open-weight Gemma 4 12B (IT) (June 2026), the Gemma 4 family's 12B Unified model with the same multimodal text, audio and image support as the E2B and E4B variants. Provider: Google. Released 2026-06-03. Access: Open.
Aggregate Rating 1626 ± 12, rank #372 of 1661 rated models, from 333 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AfriEconQA - Factoid | 71.3 | Token F1 (%; the factoid questions (exact numbers, dates and | 100 |
| SEA-HELM (Filipino) - NLU | 76.4 | Normalized Score | 98.2 |
| SEA-HELM (Filipino) - SEA-IFEval | 93.6 | Normalized Score | 98.2 |
| SEA-HELM (Burmese) - NLG | 56.23 | Normalized Score | 96.5 |
| SEA-HELM (Burmese) - Sentiment Analysis | 47.56 | Normalized Score | 96.5 |
| SEA-HELM (English) - LiveCodeBench v6 | 74.24 | Normalized Score | 96.5 |
| SEA-HELM (Filipino) - Belebele QA | 87.88 | Normalized Score | 96.5 |
| SEA-HELM (Filipino) - Causal Reasoning | 94.18 | Normalized Score | 96.5 |
| SEA-HELM (Filipino) - Summarization | 25.81 | Normalized Score | 96.5 |
| SEA-HELM (Indonesian) - Metaphor Understanding | 85.23 | Normalized Score | 96.5 |
| SEA-HELM (Malay) - SEA-IFEval | 90.12 | Normalized Score | 96.5 |
| SEA-HELM (Malay) - Sentiment Analysis | 52.86 | Normalized Score | 96.5 |
Interactive version: theaggregate.ai/model?slug=gemma-4-12b-it · How It Works · Data refreshed daily, snapshot 2026-10-23.