MODEL PROFILE / EDITION 2026.09
Qwen3.8 Max 0902
Cost-conscious teams that want frontier frontend quality plus native chart/screenshot understanding
Balanced model evidence score; rank 6. No rank delta is shown because there is no prior numeric snapshot.
Best for
Cost-conscious teams that want frontier frontend quality plus native chart/screenshot understanding
Price & access
Singapore/international $2 in / $6 out per 1M (0<token<=1M tier); other regions $1.65/$4.951; Alibaba's own docs list CNY 14.988/44.965 (Singapore) and CNY 12/36 elsewhere; batch calls 50% of real-time; context-cache hits ~10% of standard input
Evidence summary
First-party Alibaba docs give exact context, modality and price; SWE-bench Pro and Arena are third-party. But AA has no 0902 row, the Arena interval is wide, and TechNode's 1,691 differs from Arena's 1700 snapshot.
Edition status
No prior snapshot. Tracked in the September 2026 baseline.
Strengths
- Rank 4 on Arena WebDev Frontend (1700) - best non-US frontend score
- SWE-bench Pro 67.7% at $2/$6 is exceptional price-performance
- Native image and video input with a real 1M context and 131K output
- Very low TTFT (2.38 s) for a frontier-scale model
Tradeoffs
- Only 1,523 Arena votes, so the +/-18 interval overlaps ranks 3-7
- AA has no 0902-specific row - its 47 index is the base snapshot
- 40.2 tok/s output speed is slow
- Hosted-only; the open-weight Qwen3.8 sibling is a different artifact
Score profile
visual 85 · code 84 · agentic 82 · debug 78 · context 92 · speed 66 · value 80
Linked evidence
Evidence source 1 ↗Evidence source 2 ↗Evidence source 3 ↗Evidence source 4 ↗Evidence source 5 ↗Evidence source 6 ↗Evidence source 7 ↗Use the same published data, filters, and side-by-side metrics.