- Published
- Sep 16, 2026, 06:28 AM UTC
- Question pack
- coding-fast-v4.12
- Tested
- 1 measured configuration
- Batch
- evaluation-497e806be8b8470345169c1aad6752d74a0c1ce9e195a9eb7fcc02b628701c77
Measured configurations
| Configuration | Rank | Score | Elapsed | Reference cost | Route |
|---|---|---|---|---|---|
| Qwen 3.8 / Max | #28 | 60/100 | 47m 48s | $1.07 | Custom endpoint |
Five-question profile
| Question | Qwen 3.8 / Max |
|---|---|
| Black-box Regression Audit | 12/20 |
| Retry Planner Counterexamples | 17/20 |
| CI Adversarial Audit | 8/20 |
| Transaction Regression Design | 14/20 |
| Cache Propagation Certificate | 9/20 |
Why the raw configuration matters
The measured tier is encoded in the Qwen model identity while the endpoint can expose a default effort field. ModelDial presents it as Max but retains the raw configuration so the result stays reproducible.
If only one Qwen 3.8 configuration is present in the selected batch, the page can compare it with other configurations but cannot infer a Qwen effort curve.