Qwen 3.5 Omni Plus: benchmark results
Alibaba's flagship Qwen 3.5 Omni tier, natively omni-modal with speech output. Provider: Alibaba. Released 2026-03-30. Access: API.
Aggregate Rating 1774 ± 9, rank #84 of 1656 rated models, from 245 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LongAudioSpan - Accuracy | 74.6 | Accuracy (%, mean over S/M/L tiers) | 100 |
| LongAudioSpan - Chain | 29 | Chain score (%, mean over S/M/L tiers) | 100 |
| ODU-Bench (Audio-Visual) - Key-Point Coverage | 65.8 | Key-point hit rate (%; share of reference key points on stru | 100 |
| Omni2Web - Instruction Recovery | 49.14 | Instruction Recovery Score (%; Track B: mean of target F1 an | 100 |
| Omni2Web - Instruction Recovery Target F1 | 49.99 | Target F1 (%; Track B component: recovered edit targets matc | 100 |
| Omni2Web - Instruction Utility | 51.08 | Edit Fidelity Score (%; Track C: the model's Track B instruc | 100 |
| PitchBench - All Note Onsets and Offsets | 30 | Accuracy (%; report every onset and offset in a sequence) | 100 |
| PitchBench - Chord Pitch Enumeration | 15.6 | Accuracy (%; list every pitch of a chord) | 100 |
| PitchBench - Continuous Pitch Trajectory | 51.9 | Accuracy (%; describe the shape of a pitch glide) | 100 |
| PitchBench - Discrete Melodic Contour | 15.6 | Accuracy (%; give the up/down contour of a stepwise sequence | 100 |
| PitchBench - Hidden Tone in Silence | 90 | Accuracy (%; name the pitch of one tone hidden in 60 s of si | 100 |
| PitchBench - Higher or Lower | 65.2 | Accuracy (%; say which of two tones is higher) | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-omni-plus · How It Works · Data refreshed daily, snapshot 2026-10-20.