Gemma 4 31B FP8: benchmark results
Provider: Google. Released 2026-04-02.
Aggregate Rating 1734 ± 48, rank #225 of 10916 rated models, from 13 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EYT-Bench (PersonaMem-v2) - Final Intent Completion | 90 | Final-intent completion rate (%; share of dialogues in which | 100 |
| EYT-Bench (PersonaMem-v2) - Emotion | 71.6 | Per-turn emotion accuracy (%; the target model first predict | 87.5 |
| EYT-Bench (PersonaMem-v2) - Explicit Intent | 76 | Per-turn explicit-intent accuracy (%; the target model first | 87.5 |
| EYT-Bench (PersonaMem-v2) - Latent Intent | 77 | Per-turn latent-intent accuracy (%; the target model first p | 87.5 |
| EYT-Bench (PersonaMem-v2) - Persona Alignment | 94.8 | Persona alignment (style, value, culture, memory, security) | 78.1 |
| EYT-Bench (Nemotron-USA) - Explicit Intent | 39.9 | Per-turn explicit-intent accuracy (%; the target model first | 75 |
| EYT-Bench (Nemotron-USA) - Latent Intent | 41.9 | Per-turn latent-intent accuracy (%; the target model first p | 75 |
| EYT-Bench (Nemotron-USA) - Emotion | 33.7 | Per-turn emotion accuracy (%; the target model first predict | 71.9 |
| EYT-Bench (PersonaMem-v2) - Empathy | 69.5 | Empathy (recognition, attribution, resonance, response, supp | 68.8 |
| EYT-Bench (PersonaMem-v2) - Anthropomorphic Interaction | 88.7 | Anthropomorphic interaction (colloquial, emotional, flow, fl | 56.2 |
| EYT-Bench (Nemotron-USA) - Empathy | 72.3 | Empathy (recognition, attribution, resonance, response, supp | 53.1 |
| GIM | 0.1 | IRT ability (theta) | 47.8 |
Interactive version: theaggregate.ai/model?slug=gemma-4-31b-fp8 · How It Works · Data refreshed daily, snapshot 2026-11-08.