Ling-flash-2.0: benchmark results
InclusionAI's speed-oriented open MoE (100B total/6.1B active), matching ~40B dense models at higher speed (September 2025). Provider: InclusionAI. Released 2025-09-17. Access: Open.
Aggregate Rating 1568 ± 14, rank #550 of 1656 rated models, from 145 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SciEval - Astronomy | 48.12 | Astronomy (%) | 94.4 |
| Medmarks - MedHallu Medium | 72.2 | Score (%) | 87.5 |
| Medmarks - MedHallu Hard | 54.44 | Score (%) | 86.4 |
| Medmarks - MedHallu Easy | 76.47 | Score (%) | 83.5 |
| SABER - Contextual Warnings | 81.3 | Harmful safety-violation rate (HSR, %) over effective runs o | 75 |
| AA Omniscience - Software Engineering (SWE) - PHP | 28.57 | Accuracy (%) | 71 |
| JP-TL-Bench - Japanese to English | 9 | LT Score (0-10; anchored pairwise Bradley-Terry vs 20 Base S | 68.3 |
| JP-TL-Bench - Japanese to English - Hard | 9.19 | LT Score (0-10; Hard items only; anchored pairwise Bradley-T | 68.3 |
| JP-TL-Bench - Japanese to English - Easy | 8.85 | LT Score (0-10; Easy items only; anchored pairwise Bradley-T | 66.7 |
| SciEval - Earth Sciences | 69.55 | Earth Sciences (%) | 66.7 |
| Medmarks - PubHealthBench Reviewed | 86.49 | Score (%) | 66.5 |
| Medmarks - PubMedQA | 76.47 | Score (%) | 65.9 |
Interactive version: theaggregate.ai/model?slug=ling-flash-2-0 · How It Works · Data refreshed daily, snapshot 2026-10-20.