Qwen3.8-Max
Alibaba · Undisclosed (MoE, trillion-scale total)
The flagship that changed who is #1
Released 9 October 2026 and the first model to take #1 on this tier list since it launched: composite 95.4, edging Doubao Seed 2.0 Pro by 0.2. It widens Qwen3.5’s document lead (97) while closing the gaps that used to trail Doubao — video timing (93) and GUI agents (95) — with 256K native context expandable to 1M.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- Qwen3.8-Max95.4
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 2026-10 | 91.8 | Official; not yet independently reproduced |
| OmniDocBench v1.5 | 0.089 | Edit distance, lower is better |
| 上下文 | 1M | 256K native, expandable |
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 | 97 · Best layout and field alignment today | |
| Chart data extraction | 96 · Strong on multi-series and cross-page | |
| 30-min video QA | 92 · Gap closed; still just behind Doubao | |
| Mobile GUI control | 95 · Cross-app planning now top tier | |
| Chinese cultural context | 95 · Strong on official and academic text, weaker on memes |
Strengths and limitations
Strengths
- Documents and charts both 96+: cross-page tables, multi-figure comparison and unit conversion are the strongest publicly visible today.
- 1M context with native vision reads a whole technical manual plus screenshots without chunking.
- GUI agents jumped from 91 on Qwen3.5 to 95; cross-app long-horizon planning enters the top tier for the first time.
Limitations
- At ¥18/1M output tokens it is one of the three priciest here; high-frequency image analysis gets a heavy bill.
- Evidence grade B: under two weeks old, with no independent reproduction and no 30-day stability record.
- Chinese everyday priors still trail Doubao; memes and regional dishes are a notch weaker zero-shot.
Price and real cost
List price
¥18/M tokens
≈ ¥18 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥10.9 (assuming input at 25% of output price).
Use your volumeAbout Qwen3.8-Max
Is Qwen3.8-Max open source?
Licence: Closed commercial API (smaller siblings open). No public weights; access is via vendor API or an on-prem licence.
How much does Qwen3.8-Max cost?
≈ ¥18 / 1M output tokens. It ranks #1 with a composite of 95.4. Real bills depend on resolution, tiling, caching and volume discounts.
What is Qwen3.8-Max best for?
Long-document, chart-dense enterprise knowledge bases; Whole-manual QA that needs 1M context; Cross-app GUI agents and automation flows。
Why does Qwen3.8-Max score 95.4?
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 Visual perception & grounding; 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.