Qwen3.5-397B-A17B
Alibaba · 397B total / 17B active (MoE)
The multimodal flagship that behaves like infrastructure
SuperCLUE-VLM 88.50, strongest in document parsing and multilingual OCR. 32-language OCR, 256K context and a full open-weight matrix make it the easiest enterprise bet.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- Qwen3.5-397B-A17B94.6
Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.
Public benchmarks
as reported, not converted| Benchmark / metric | Value | Note |
|---|---|---|
| SuperCLUE-VLM 2026-04 | 88.50 | Qwen3.5-397B-A17B-Thinking |
| OCR 语言覆盖 | 32 | Qwen3-VL family, official |
| 上下文 | 256K | Expandable to 1M |
Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.
Scenario mapping
five stress tests, one scale| Scenario | Score | Observation |
|---|---|---|
| Chinese receipt OCR | 96 · Good layout fidelity and field alignment | |
| Chart data extraction | 95 · Strong cross-page comparison | |
| 30-min video QA | 91 · Slightly behind Doubao on long video | |
| Mobile GUI control | 91 · Stable UI element grounding | |
| Chinese cultural context | 94 · Solid local knowledge |
Strengths and limitations
Strengths
- Best-balanced document and 32-language OCR; table structure is stable.
- A full 8B-to-397B matrix, one API from edge to cloud.
- Richest open ecosystem and fine-tuning tooling; cheapest community reproduction.
Limitations
- The flagship is closed; only small and mid sizes are truly self-hostable.
- Details late in very long documents can slip; chunking plus retrieval is advised.
- Large capability gaps between sizes make mis-selection easy.
Price and real cost
List price
¥12/M tokens
≈ ¥12 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥7.3 (assuming input at 25% of output price).
Use your volumeAbout Qwen3.5-397B-A17B
Is Qwen3.5-397B-A17B open source?
Licence: Smaller sizes open (Apache-2.0); flagship closed. No public weights; access is via vendor API or an on-prem licence.
How much does Qwen3.5-397B-A17B cost?
≈ ¥12 / 1M output tokens. It ranks #3 with a composite of 94.6. Real bills depend on resolution, tiling, caching and volume discounts.
What is Qwen3.5-397B-A17B best for?
Enterprise document hubs and contract review; Multilingual product data extraction; On-prem public-sector projects。
Why does Qwen3.5-397B-A17B score 94.6?
The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Ecosystem, cost & availability; its weakest is Generation & editing.
Personal ranking · Not an official score
Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.