Fanar-2-27B-Instruct: benchmark results
Provider: Other. Released 2026-03-16. Access: Open.
Unified ELO 1636 ± 55, rank #505 of 2656 rated models, from 24 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| QIMMA - MedArabiQ MCQ | 48.24 | Accuracy (%, log-likelihood multiple choice) | 86.6 |
| QIMMA - GAT | 51.52 | Accuracy (%, log-likelihood multiple choice) | 82.3 |
| QIMMA - Overall | 60.64 | Mean score across 14 benchmarks (%) | 80.5 |
| QIMMA - MedAraBench | 42.31 | Accuracy (%, log-likelihood multiple choice) | 75.6 |
| QIMMA - AraDiCE-Culture | 67.78 | Accuracy (%, log-likelihood multiple choice) | 73.8 |
| QIMMA - ArabCulture | 58.14 | Accuracy (%, log-likelihood multiple choice) | 70.7 |
| QIMMA - FannOrFlop | 50.49 | F1 (%) | 67.1 |
| QIMMA - 3LM STEM | 80.64 | Accuracy (%, log-likelihood multiple choice) | 65.9 |
| QIMMA - ArabicMMLU | 65.96 | Accuracy (%, log-likelihood multiple choice) | 65.9 |
| QIMMA - 3LM HumanEval+ | 57.32 | pass@1 (%) | 62.2 |
| QIMMA - PALMX | 70.03 | Accuracy (%, log-likelihood multiple choice) | 61 |
| QIMMA - 3LM MBPP+ | 56.35 | pass@1 (%) | 60.4 |
Interactive version: theaggregate.ai/model?slug=fanar-2-27b-instruct · How It Works · Data refreshed daily, snapshot 2026-09-19.