Qwen3.5 Omni Flash: benchmark results
Alibaba's low-latency Flash tier of the omni-modal Qwen3.5-Omni family: native text/image/audio/video input with real-time speech output (March 2026). Provider: Alibaba. Released 2026-03-30. Access: API.
Aggregate Rating 1708 ± 13, rank #177 of 1656 rated models, from 122 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PitchBench - Slightly Out-of-Tune Pitch | 30 | Accuracy (%; name the nearest in-tune pitch of a detuned ton | 100 |
| VocalCoachBench - Triplet Ranking | 72.4 | Pairwise accuracy (%; for three performances of the same son | 88.9 |
| LongAudioSpan - Accuracy | 68.1 | Accuracy (%, mean over S/M/L tiers) | 85.7 |
| PitchBench - All Note Onsets and Offsets | 2.5 | Accuracy (%; report every onset and offset in a sequence) | 80 |
| PitchBench - Chord Pitch Enumeration | 12.8 | Accuracy (%; list every pitch of a chord) | 80 |
| PitchBench - Hidden Tone in Silence | 75.6 | Accuracy (%; name the pitch of one tone hidden in 60 s of si | 80 |
| PitchBench - Named Note Onset and Offset | 2.5 | Accuracy (%; report the onset and offset of a named note amo | 80 |
| PitchBench - Pitch Identification | 75.1 | Accuracy (%; name the pitch of a single sustained tone) | 80 |
| PitchBench - Pitch at a Queried Timestamp | 56.7 | Accuracy (%; name the note sounding at a given time in a seq | 80 |
| PitchBench - Pitch under Audio Effects | 55.8 | Accuracy (%; name the pitch after filtering, bitcrushing, sa | 80 |
| PitchBench - Pitch under Background Noise | 22.9 | Accuracy (%; name the pitch against background sounds at fou | 80 |
| PitchBench - Pitch under Duration Variation | 61 | Accuracy (%; name the pitch of tones from 50 ms to 60 s) | 80 |
Interactive version: theaggregate.ai/model?slug=qwen3-5-omni-flash · How It Works · Data refreshed daily, snapshot 2026-10-20.