Falcon3-1B-Instruct: benchmark results
TII's Falcon3-1B-Instruct is an 18-layer decoder-only Transformer instruction model using grouped-query attention and a publisher-supported 8K context. It is pruned and healed from larger Falcon models using 80 billion tokens. Instruction post-training uses 1.2 million samples. Provider: TII. Released 2024-12-17. Access: Open.
Aggregate Rating 1217 ± 24, rank #9504 of 10916 rated models, from 98 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MEDIC Benchmark | 69.39 | MEDIC Public Table Average | 73.7 |
| Open LLM Leaderboard - IFEval | 55.57 | Score | 67.2 |
| Open LLM Leaderboard - MuSR | 41.89 | Score | 59.5 |
| Open LLM Leaderboard - MATH Level 5 | 6.34 | Score | 34.5 |
| Open Portuguese LLM - FaQuAD NLI | 51.95 | Macro F1 (%) | 33.4 |
| EuroEval Portuguese NLU - MultiWikiQA PT | 52.49 | Character F1 (%) | 29.7 |
| Open Portuguese LLM - ASSIN2 RTE | 76.08 | Macro F1 (%) | 27.7 |
| Open Portuguese LLM - BLUEX | 31.57 | Accuracy (%) | 27.6 |
| Open Portuguese LLM - ENEM | 34.57 | Accuracy (%) | 27.3 |
| Open Portuguese LLM - OAB Exams | 30.84 | Accuracy (%) | 27 |
| EuroEval Portuguese NLU - ScaLA PT | 2.08 | Matthews correlation coefficient (%) | 25.2 |
| Open LLM Leaderboard - BBH | 37.45 | Score | 21.5 |
Interactive version: theaggregate.ai/model?slug=falcon3-1b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.