recurrentgemma-2B-it: benchmark results
Google DeepMind's RecurrentGemma 2B-it, the instruction-tuned variant using the Griffin architecture of linear recurrences plus local attention for lower memory use and faster long-sequence inference than Gemma. Provider: Google. Released 2024-04-11. Access: Open.
Aggregate Rating 1223 ± 30, rank #1542 of 1635 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 10.08 | Accuracy (%) (5-shot) | 31 |
| Open Portuguese LLM - ASSIN2 RTE | 79.42 | Macro F1 (%) | 30.1 |
| Open Portuguese LLM - BLUEX | 31.43 | Accuracy (%) | 27.2 |
| Open LLM Leaderboard - IFEval | 29.49 | Score | 26.7 |
| Open Portuguese LLM - ENEM | 32.68 | Accuracy (%) | 26 |
| Open Portuguese LLM - OAB Exams | 30.34 | Accuracy (%) | 25.7 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 42.81 | MC2 (%) (0-shot) | 21.6 |
| Open LLM Leaderboard v1 - MMLU | 40.87 | Accuracy (%) (5-shot) | 21.4 |
| Open LLM Leaderboard v1 - WinoGrande | 64.17 | Accuracy (%) (5-shot) | 17 |
| Open LLM Leaderboard - BBH | 33.3 | Score | 16.5 |
| Open LLM Leaderboard - MATH Level 5 | 1.96 | Score | 13.7 |
| Open LLM Leaderboard v1 - HellaSwag | 56.26 | Normalized accuracy (%) (10-shot) | 12.5 |
Interactive version: theaggregate.ai/model?slug=recurrentgemma-2b-it · How It Works · Data refreshed daily, snapshot 2026-11-05.