Gemma 4 31B (IT): benchmark results
Instruction-tuned Gemma 4 31B checkpoint. Provider: Google. Released 2026-04-02. Access: Open.
Aggregate Rating 1709 ± 4, rank #176 of 1656 rated models, from 1775 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - chinese_homophonic_puns | 1.56 | Dataset z-score | 100 |
| AGC-Bench - outline_to_story | 1.81 | Dataset z-score | 100 |
| ArmBench-LLM - POS | 80 | Accuracy (%) | 100 |
| ArtECulture - English | 59.35 | Accuracy (%; English-speaking culture labels; zero-shot pred | 100 |
| CDR-Bench - Order-Agnostic Recipes | 25.55 | Recipe success RS@3 (%) on the order-agnostic mapper-recipe | 100 |
| CSTutorBench | 89 | Overall rubric score (%; points earned out of the maximum on | 100 |
| CSTutorBench - Accuracy | 94 | Accuracy criterion (%; no incorrect claims about block behav | 100 |
| Codeforces ELO (Google Gemma 4 report) | 2150 | Codeforces ELO, Google Gemma 4 instruction-tuned report (sel | 100 |
| EuroEval Danish Summarization - Nordjylland News | 38.54 | chrF3++ (%) | 100 |
| EuroEval French NLU | 79.51 | NLU Average Score (%) | 100 |
| EuroEval Lithuanian NLU - Atsiliepimai | 50.67 | Matthews correlation coefficient (%) | 100 |
| EuroEval Polish NLU | 68.96 | NLU Average Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=gemma-4-31b-it · How It Works · Data refreshed daily, snapshot 2026-10-20.