On 2026-10-15, the YZ Index Smoke quick test covered 13 models, with Gemini 2.5 Pro ranking first for the day at 95.91. Smoke is a daily 10-question quick test, suitable for observing short-term signals; it is not equivalent to Full weekly ranking conclusions.
This Smoke evaluation covered 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 more suitable as monitoring signals than as long-term conclusions about model capability.
Daily Ranking
| Ranking | Model | Main Leaderboard | Code Execution | Material Constraints | Integrity |
|---|---|---|---|---|---|
| #1 | Gemini 2.5 Pro | 95.91 | 97.8 | 93.6 | pass |
| #2 | Claude Opus 4.7 | 93.54 | 97 | 89.3 | pass |
| #3 | Claude Sonnet 4.6 | 90.71 | 93 | 87.9 | pass |
| #4 | Doubao Pro | 89.44 | 90.7 | 87.9 | pass |
| #5 | GPT-5.5 | 87.86 | 92 | 82.8 | pass |
| #6 | GPT-6 Sol | 87.86 | 92 | 82.8 | pass |
| #7 | Grok 4 | 87.05 | 95.1 | 77.2 | pass |
| #8 | GPT-6 Luna | 85.97 | 92 | 78.6 | pass |
| #9 | DeepSeek V4 Pro | 84.6 | 72 | 100 | pass |
| #10 | GPT-o3 | 84.6 | 72 | 100 | pass |
| #11 | GPT-6 Astra | 77.04 | 67 | 89.3 | pass |
| #12 | GPT-6.1 Sol | 77.04 | 67 | 89.3 | pass |
| #13 | Gemini 3.1 Pro | 63.29 | 42 | 89.3 | pass |
Data Interpretation
In today's YZ Index Smoke quick test, Gemini 2.5 Pro ranked first on the main leaderboard with 95.91, combining a balanced high score of 97.8 in code execution and 93.6 in material constraints. Claude Opus 4.7 followed closely, with a main leaderboard score of 93.54 corresponding to 97 in code execution and 89.3 in material constraints, also reflecting stable performance with a code-execution emphasis. DeepSeek V4 Pro and GPT-o3 tied at 84.6 on the main leaderboard; both scored 72 in code execution and 100 in material constraints, showing a complementary structure of very strong material constraints but weaker code execution. Doubao Pro and Claude Sonnet 4.6 maintained a mid-to-high balance in the ranges of 90.7 and 93 in code execution and 87.9 in material constraints.
In terms of anomalous signals, DeepSeek V4 Pro and GPT-o3 each showed a -22.5 point change in code execution; Gemini 3.1 Pro fell 12.6 points on the main leaderboard and 22.5 points in code execution; GPT-6.1 Sol fell 11.3 points on the main leaderboard and 22.5 points in code execution; and GPT-6 Astra fell 10.2 points on the main leaderboard. These single-day fluctuations may stem from question sampling variability or may reflect genuine signs of degradation, requiring follow-up runs for confirmation. GLM-4.6 and Qwen3 Max had incomplete data due to API failures and did not participate in this period's ranking.
Overall, leading models mostly relied on code execution scores above 97 to drive their main leaderboard results, while models with 100 in material constraints fell back to the 84.6 range on the main leaderboard when code execution was 72, showing the direct impact of the strength/weakness combination across the two dimensions on final rankings. Under the single-day signals of a small-sample Smoke test, all interpretations remain restrained, with no long-term inferences.
Key Changes
- Gemini 2.5 Pro: main leaderboard up 19.6 points, code execution +32.5 points
- Gemini 3.1 Pro: main leaderboard down 12.6 points, code execution -22.5 points
- Claude Sonnet 4.6: main leaderboard up 12.1 points, code execution +23.5 points
- GPT-6 Luna: main leaderboard up 11.4 points, code execution +27.5 points, material constraints -8.2 points
- GPT-6.1 Sol: main leaderboard down 11.3 points, code execution -22.5 points
Signals to Watch
- DeepSeek V4 Pro: code execution plunged -22.5 points
- GPT-o3: code execution plunged -22.5 points
- GPT-6 Astra: main leaderboard plunged -10.2 points
- GPT-6.1 Sol: main leaderboard plunged -11.3 points
- Gemini 3.1 Pro: main leaderboard plunged -12.6 points
- GLM-4.6: incomplete data (missing execution dimension, API failure/timeout), has entered automatic rerun, not included in this period's ranking
- Qwen3 Max: incomplete data (missing multiple evaluation dimensions, API failure/timeout), has entered automatic rerun, not included in this period's ranking
When reading this kind of Smoke brief, the focus should be on two questions: first, whether a model has repeatedly exposed the same type of weakness over multiple days; second, whether its integrity rating has moved from pass to warn or fail. Large single-day changes in execution or constraint scores may come from question sampling or may be early signals of genuine degradation, requiring follow-up runs for confirmation.
Data source: YZ Index | Run #380 | View raw data
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接