Falcon3-10B-Instruct: benchmark results
TII's Falcon3-10B-Instruct is a 40-layer decoder-only Transformer instruction model using grouped-query attention and a publisher-supported 32K context. Its Falcon3-7B-based depth expansion uses two trillion continued-pretraining tokens. Instruction post-training uses 1.2 million samples. Provider: TII. Released 2024-12-17. Access: Open.
Aggregate Rating 1417 ± 16, rank #1162 of 1661 rated models, from 233 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - IFEval | 78.17 | Score | 94.4 |
| Open Japanese LLM - Xlsum JA Rouge1 | 35.03 | Score (%) | 93.3 |
| Open Portuguese LLM - FaQuAD NLI | 81.69 | Macro F1 (%) | 92.9 |
| Open Portuguese LLM - ASSIN2 RTE | 93.71 | Macro F1 (%) | 91.4 |
| MEDIC Benchmark | 81.03 | MEDIC Public Table Average | 87.4 |
| Open Portuguese LLM - OAB Exams | 54.26 | Accuracy (%) | 85.7 |
| Open Japanese LLM - Xlsum JA Rouge2 | 11.3 | Score (%) | 85.5 |
| Open LLM Leaderboard - BBH | 61.7 | Score | 85.4 |
| Open Japanese LLM - SUM | 11.28 | Score (%) | 85.3 |
| French LLM Leaderboard - IFEval FR | 63.8 | Score (%) | 85.2 |
| Open Japanese LLM - Xlsum JA Bert Score JA F1 | 70.34 | Score (%) | 84.5 |
| Open Portuguese LLM - ENEM | 72.57 | Accuracy (%) | 83.5 |
Interactive version: theaggregate.ai/model?slug=falcon3-10b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-23.