Best cell per row highlighted. Null means undisclosed — never counted as zero.
| Provider | Alibaba Cloud | Alibaba Cloud |
| Family / tier | Qwen3.8 · Max | Qwen3.8 · Large |
| Status | GA | GA |
| Released | 2026-08-03 | 2026-08-14 |
| Knowledge cutoff | — | — |
| Context window | 1M | 1M |
| Max output tokens | 131K | 131K |
| Input modalities | text, image, video | text, image, video |
| Reasoning mode | optional | optional |
| Open weights | Yes | Yes |
| Input | $2 | $0.5 |
| Output | $6 | $3 |
| Cached input | $0.17 | $0.05 |
| Batch input | — | — |
| Free tier | Yes | Yes |
| Output speed (tok/s) | 39.1 | 46.4 |
| Time to first token (s) | 3.02 | 3.94 |
| Latency tier | slow | slow |
| Trains on inputs | — | — |
| License | custom-qwen3.8-max-license | Apache-2.0 |
| Clouds | — | Alibaba Cloud Model Studio |
| Compliance | — | — |
| Unified score | 8.6 | 8.5 |
| Decision Maker | 8.5 | 9 |
| Domain Strategist | 9 | 9 |
| Finance Lead | 8.5 | 8.5 |
| Domain Practitioner | 8.5 | 9 |
| Power User | 8 | 7.5 |
| Skeptic | 7.5 | 8 |
| Value score | 8.8 | 9 |
| GPQA Diamond | 92.6 | 90.5 |
| LiveCodeBench | — | 90.3 |
| Terminal-Bench | 86.6 | 79.8 |
| MRCR Long Context | 92.9 | — |
| Humanity's Last Exam | 43.6 | 33.9 |
| LMArena Elo | 1481 | 1437 |
| LMArena Coding Elo | 1522 | 1499 |
| Artificial Analysis Index | 40 | 34 |
Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.