Llama-3.1-Tulu-3-8B: benchmark results

Allen AI's fully open Tulu 3 post-train of Llama 3.1 8B, finishing with RLVR after SFT and DPO for strong math and instruction following (November 2024). Provider: Allen AI. Released 2024-11-21. Access: Open.

Aggregate Rating 1480 ± 22, rank #849 of 1656 rated models, from 191 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AbstentionBench - underspecified context - MediQ - Precision100Abstention Precision (%)100
Open LLM Leaderboard - IFEval82.67Score99.1
AbstentionBench - underspecified context - UMWP - Precision100Abstention Precision (%)97.2
AbstentionBench - stale - FreshQA - F151.57Abstention F1 (%)94.7
AbstentionBench - answer unknown - KUQ/Future unknowns - F182.56Abstention F1 (%)89.5
AbstentionBench - answer unknown - KUQ/Future unknowns - Precision94.47Abstention Precision (%)89.5
AbstentionBench - answer unknown - KUQ/Unsolved problems - Precision91.62Abstention Precision (%)89.5
AbstentionBench - stale - CoCoNot/Temporal - F192.75Abstention F1 (%)89.5
AbstentionBench - stale - CoCoNot/Temporal - Recall86.49Abstention Recall (%)89.5
AbstentionBench - stale - FreshQA - Recall87.23Abstention Recall (%)89.5
AbstentionBench - underspecified context - Musique - Precision98.73Abstention Precision (%)89.5
AbstentionBench - underspecified context - QASPER - F144.31Abstention F1 (%)89.5

Interactive version: theaggregate.ai/model?slug=llama-3-1-tulu-3-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.