Ling-3.0-flash-VL: benchmark results
InclusionAI's open-weight Ling-3.0-flash-VL (September 2026), a native multimodal model built on Ling-3.0-flash with 124B total and 5.5B active parameters, adding image and video understanding via a ViT visual encoder. Provider: InclusionAI. Released 2026-09-04. Access: Open.
Aggregate Rating 1704 ± 19, rank #186 of 1635 rated models, from 173 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - jaster (0-shot) - MAWPS | 99 | Exact match (%) | 98.6 |
| AA-Omniscience Hallucination Rate | 22.05 | Hallucination Rate (%) | 97.6 |
| Nejumi 4 - jaster (2-shot) - MAWPS | 99 | Exact match (%) | 96.8 |
| Nejumi 4 - jaster (0-shot) - JSQuAD | 81.58 | Character F1 (x100) | 92.9 |
| Nejumi 4 - MT-Bench (Japanese) - Humanities | 100 | Judge rating (1-10, x10) | 90.8 |
| Nejumi 4 - ALT - Bias | 94.45 | Score (%) | 90.2 |
| Nejumi 4 - JHumanEval | 42.23 | Sandbox pass rate (%) | 84.4 |
| Nejumi 4 - GLP - Applied Language | 89.63 | Score (%) | 82.3 |
| Nejumi 4 - jaster (2-shot) - JaMP | 79 | Exact match (%) | 81.9 |
| Nejumi 4 - ARC-AGI-2 | 50 | Accuracy (%) | 81.6 |
| AA GPQA Diamond | 86.16 | Accuracy (%) | 81 |
| AA Long Context Reasoning | 78.33 | Accuracy (%) | 80.3 |
Interactive version: theaggregate.ai/model?slug=ling-3-0-flash-vl · How It Works · Data refreshed daily, snapshot 2026-10-15.