Qwen 3.5 397B A17B (Non-reasoning) Benchmark Results

Qwen 3.5 397B A17B (Non-reasoning): benchmark results

Qwen 3.5 397B A17B evaluated with reasoning disabled. Provider: Alibaba. Released 2026-02-16. Access: Open.

Aggregate Rating 1651 ± 9, rank #518 of 2168 rated models, from 197 benchmark results.

Benchmark	Score	Metric	Percentile
Chronicles-OCR - Mature Text Parsing	0.73	Paragraph-level normalized edit similarity (0-1: one minus t	100
EPIC-Bench - Ground Detection	51.94	Navigation score (0-100) on ground detection (localizing tra	100
IndustryBench - Russian	2.19	Final (SV) score (out of 3): mean 0-3 rubric score from a Qw	100
LongDocBench - Source Relationship Recovery	85	Source relationship similarity (%; the model reads fixed Tex	100
PerSo - Behavioral	76	Accuracy (%)	100
Chronicles-OCR - Archaic Text Parsing	0.22	Paragraph-level normalized edit similarity (0-1: one minus t	96.4
EPIC-Bench - Embodied Compositional Attributes	43.45	Target localization score (0-100) on the embodied compositio	94.3
Chronicles-OCR - Archaic Character Spotting	9.7	Text-spotting H-mean (0-100) of predicted character boxes wi	92.9
LiT (Lost in Translation) - Humanities Abstracts	71.2	MQM>=80 rate (%): share of round-trip translations whose bac	92.9
PerSo - Belief	80.5	Accuracy (%)	92.9
VAMPS - Persian	88.5	Accuracy (%) on the original Persian versions of the 218 Kon	91.2
IndustryBench - Safety and Compliance	2.37	Final (SV) score (out of 3): mean 0-3 rubric score from a Qw	90.6