Qwen3.8-Flash
Alibaba · Undisclosed (MoE, ~12B active)
An A-tier ticket for ¥1.2
The fast sibling shipped the same day as Qwen3.8-Max. Composite 87.5 puts it in tier A at one-fifteenth of Max’s price — the first time a Flash-class model reaches A on this board, which redraws the price-performance Pareto frontier. Its ecosystem score of 97 is among the highest here: open weights, Apache-2.0, 256K context, one API from edge to cloud.
Last updated Oct 18, 2026Editorial scoring
Bottom line
Eight-dimension breakdown
editorial mapping · 0–100- Qwen3.8-Flash87.5
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 | 84.2 | #1 in its price class (official) |
| 输出单价 | ¥1.2 | ≈ 1/15 of Max |
| 上下文 | 256K | Highest in class |
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 | 90 · Near-flagship on clean receipts | |
| Chart data extraction | 85 · Fine single-figure; drops on cross-page | |
| 30-min video QA | 80 · Noticeable event-localization drift | |
| Mobile GUI control | 86 · Steady single steps; long tasks need rollback | |
| Chinese cultural context | 88 · Common memes fine, obscure allusions not |
Strengths and limitations
Strengths
- ¥1.2/1M output tokens plus a tier-A composite leaves almost nothing to beat on value.
- Apache-2.0 open weights, self-hostable, and shares an API with Max — enabling a Flash-triage plus Max-review relay.
- 256K context is unmatched at this price, so long-document triage needs less chunking.
Limitations
- Video timing is only 82; event localization drifts on long clips, so it should not carry video QA.
- Generation and editing at 80, with Chinese glyph error rates a tier above the flagship.
- It gets hard cases confidently wrong and lacks Max’s self-checking, so sampled review is mandatory.
Price and real cost
List price
¥1.2/M tokens
≈ ¥1.2 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥0.7 (assuming input at 25% of output price).
Use your volumeAbout Qwen3.8-Flash
Is Qwen3.8-Flash open source?
Licence: Open weights (Apache-2.0). Weights are available, but check the exact terms and filing requirements before commercial use.
How much does Qwen3.8-Flash cost?
≈ ¥1.2 / 1M output tokens. It ranks #13 with a composite of 87.5. Real bills depend on resolution, tiling, caching and volume discounts.
What is Qwen3.8-Flash best for?
High-volume product tagging and moderation; Batch receipt structuring with flagship sampling; Edge or low-cost on-prem deployment。
Why does Qwen3.8-Flash score 87.5?
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.