The 2026-11-13 YZ Index Smoke quick test covered 13 models, with Claude Opus 4.7 taking first place for the day at 94.91 points. Smoke is a daily 10-question quick test, suited to observing short-term signals and not equivalent to the conclusions of the weekly Full leaderboard.
This Smoke evaluation covers only two main leaderboard dimensions: code execution and material constraints. The main leaderboard formula is 0.55 × code execution + 0.45 × material constraints. Because the daily sample size is small, single-day scores are better treated as monitoring signals rather than as a long-term verdict on model capability.
Daily rankings
| Rank | Model | Main leaderboard | Code execution | Material constraints | Integrity |
|---|---|---|---|---|---|
| #1 | Claude Opus 4.7 | 94.91 | 99.5 | 89.3 | pass |
| #2 | GPT-6 Astra | 92.33 | 94.8 | 89.3 | pass |
| #3 | Claude Sonnet 4.6 | 82.31 | 99.5 | 61.3 | pass |
| #4 | GPT-5.5 | 81.55 | 100 | 59 | pass |
| #5 | GPT-o3 | 79.78 | 94.9 | 61.3 | pass |
| #6 | GPT-6.1 Sol | 78.69 | 94.8 | 59 | pass |
| #7 | GPT-6 Sol | 71.38 | 99.5 | 37 | pass |
| #8 | Grok 4 | 67.9 | 73.3 | 61.3 | pass |
| #9 | GPT-6 Luna | 60.39 | 93.6 | 19.8 | pass |
| #10 | Doubao Pro | 57.59 | 75 | 36.3 | pass |
| #11 | Gemini 3.1 Pro | 57.59 | 75 | 36.3 | pass |
| #12 | Gemini 2.5 Pro | 53.88 | 47.8 | 61.3 | pass |
| #13 | DeepSeek V4 Pro | 42.8 | 50 | 34 | pass |
Reading the data
Looking at the score structure, Claude Opus 4.7 ranks first with a main leaderboard score of 94.91, its 99.5 in code execution and 89.3 in material constraints forming a balanced pairing; GPT-6 Astra's main leaderboard score of 92.33 likewise rests on a combination of 94.8 in code execution and 89.3 in material constraints, indicating that the leading models maintain high levels on both metrics. Claude Sonnet 4.6 at 82.31 and GPT-5.5 at 81.55 on the main leaderboard, by contrast, post high scores of 99.5 and 100 in code execution but only 61.3 and 59 in material constraints, reflecting a structural pattern of strong code execution paired with relatively weak material constraints. GPT-o3 and GPT-6.1 Sol, at 79.78 and 78.69 on the main leaderboard, show a similar pairing, leading at 94.9 and 94.8 in code execution while sitting at 61.3 and 59 in material constraints.
GPT-6 Luna fell 28.5 points on the main leaderboard and 68.3 points in material constraints; Gemini 3.1 Pro fell 15.4 points on the main leaderboard and 35.1 points in material constraints; GPT-6 Sol fell 12.5 points on the main leaderboard but rose 5.5 points in code execution while falling 34.4 points in material constraints. These changes may stem from sampling fluctuation in the questions, or may be genuine degradation in material constraints, and require follow-up runs to confirm stability. GPT-5.5 rose 11.1 points on the main leaderboard and 25 points in code execution, while GPT-6 Astra rose 11 points on the main leaderboard and 24.5 points in material constraints; these too require multiple tests to distinguish incidental fluctuation from sustained performance.
As a small-sample, single-day signal, Smoke's current data offers only same-day observations; no anomalous signals appeared, and the reading remains restrained, avoiding conclusions about models' long-term capability.
Key changes
- GPT-6 Luna: main leaderboard down 28.5 points, material constraints -68.3 points
- Gemini 3.1 Pro: main leaderboard down 15.4 points, material constraints -35.1 points
- GPT-6 Sol: main leaderboard down 12.5 points, code execution +5.5 points, material constraints -34.4 points
- GPT-5.5: main leaderboard up 11.1 points, code execution +25 points, material constraints -5.8 points
- GPT-6 Astra: main leaderboard up 11 points, material constraints +24.5 points
Signals to watch
- No publishable anomalous signals were retained this time.
When reading this kind of Smoke brief, the focus should be on two questions: first, whether a given model exposes the same type of weakness across consecutive days; second, whether its integrity rating moves from pass to warn or fail. Large single-day swings in execution or constraint scores may come from question sampling, or may be early signals of genuine degradation, and require follow-up runs to verify.
Data source: YZ Index | Run #425 | View raw data
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接