Llama-3.1-Tulu-3-405B: benchmark results
Allen Institute for AI Tulu 3 final RLVR post-trained release based on Llama 3.1 405B; distinct from the SFT and DPO intermediate checkpoints. Provider: Allen AI. Released 2025-01-30. Access: Open.
Aggregate Rating 1551 ± 33, rank #1697 of 10916 rated models, from 11 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PLCC - Art & Entertainment | 64 | Accuracy (%) | 58.4 |
| PLCC - History | 75 | Accuracy (%) | 43.2 |
| PLCC - Culture & Tradition | 64 | Accuracy (%) | 41.6 |
| PLCC - Overall | 63.83 | Mean category accuracy (%) | 41.6 |
| AA MMLU-Pro | 71.62 | Accuracy (%) | 39.7 |
| PLCC - Geography | 71 | Accuracy (%) | 37.2 |
| PLCC - Vocabulary | 53 | Accuracy (%) | 37.2 |
| AA LiveCodeBench | 29.1 | Pass@1 (%) | 33.5 |
| PLCC - Grammar | 56 | Accuracy (%) | 32.5 |
| AA GPQA Diamond | 51.62 | Accuracy (%) | 25.7 |
| AA Humanity's Last Exam | 3.28 | Accuracy (%) | 3.2 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-tulu-3-405b · How It Works · Data refreshed daily, snapshot 2026-11-08.