Qwen3.5-397B-A17B scores 45 on the Intelligence Index with only 17B active parameters, but hallucinates 88% of the time on hard questions. That tradeoff matters when picking an open model for agents.
Key takeaways · AI-distilled
In February 2026, Artificial Analysis reported that Qwen3.5-397B-A17B combined native image and video inputs with reasoning and non-reasoning modes in one model. Earlier Qwen releases separated vision and text model lines.
Qwen3.5's February 2026 AA-Omniscience improvement came mainly from accuracy rising from 22% to 30%. Its hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → rate changed only from 90% to 88%, so the improved index score did not resolve its abstention weakness.
Artificial Analysis' February 2026 Qwen3.5 report distinguished the 262,000-tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → open model context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → from the hosted Qwen3.5-Plus variant's one-million-token context. Those deployment variants should not share an undifferentiated specification.
Terms in this piece · Glossary
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
