The 2026-10-23 YZ Index Smoke quick test covered 14 models, with Doubao Pro and GPT-6.1 Sol tied for first place that day at 87.27 points. Smoke is a daily 10-question quick test, suitable for observing short-term signals and not equivalent to Full weekly leaderboard 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 better used as monitoring signals rather than long-term conclusions about model capabilities.
Daily Ranking
| Rank | Model | Main Leaderboard | Code Execution | Material Constraints | Integrity |
|---|---|---|---|---|---|
| #1 | Doubao Pro | 87.27 | 100 | 71.7 | pass |
| #2 | GPT-6.1 Sol | 87.27 | 100 | 71.7 | pass |
| #3 | Grok 4 | 85.02 | 100 | 66.7 | pass |
| #4 | GPT-o3 | 83.37 | 97 | 66.7 | pass |
| #5 | GPT-6 Sol | 83.04 | 97.8 | 65 | pass |
| #6 | GPT-6 Astra | 81.39 | 94.8 | 65 | pass |
| #7 | Claude Sonnet 4.6 | 78.27 | 100 | 51.7 | pass |
| #8 | Claude Opus 4.7 | 78.02 | 75 | 81.7 | pass |
| #9 | GPT-6 Luna | 78.02 | 75 | 81.7 | pass |
| #10 | DeepSeek V4 Pro | 73.52 | 75 | 71.7 | pass |
| #11 | Gemini 3.1 Pro | 73.52 | 75 | 71.7 | pass |
| #12 | GPT-5.5 | 71.27 | 75 | 66.7 | pass |
| #13 | Gemini 2.5 Pro | 67.81 | 72.8 | 61.7 | pass |
| #14 | GLM-4.6 | 18.25 | 25 | 10 | fail |
Data Interpretation
Today's YZ Index Smoke quick test shows that Doubao Pro and GPT-6.1 Sol both scored 87.27 on the main leaderboard, with both at 100 for Code Execution and 71.7 for Material Constraints, reflecting a structure of strong Code Execution paired with moderate Material Constraints. Grok 4 scored 85.02 on the main leaderboard, with 100 for Code Execution and 66.7 for Material Constraints, also relying on high Code Execution to support its ranking. Claude Opus 4.7 and GPT-6 Luna both scored 78.02 on the main leaderboard, with 75 for Code Execution and 81.7 for Material Constraints, showing a pairing with stronger Material Constraints, while Claude Sonnet 4.6 scored 78.27 on the main leaderboard with 100 for Code Execution and 51.7 for Material Constraints.
GPT-5.5 fell 24.8 points on the main leaderboard, with Code Execution down 22 points and Material Constraints down 28.3 points; Gemini 2.5 Pro fell 24.7 points on the main leaderboard, with Code Execution down 27.2 points and Material Constraints down 21.6 points; Gemini 3.1 Pro fell 17.3 points on the main leaderboard, with Code Execution down 22 points and Material Constraints down 11.6 points; DeepSeek V4 Pro fell 15.6 points on the main leaderboard, with Code Execution down 18.9 points and Material Constraints down 11.6 points. These score fluctuations may come from question sampling volatility and require follow-up runs to confirm whether there is real degradation. Claude Sonnet 4.6 rose 14.1 points on the main leaderboard, with Code Execution up 31.1 points and Material Constraints down 6.6 points.
The Smoke quick test is a small-sample single-day signal. Models differ in how they pair strengths in Code Execution and Material Constraints, and movements should be interpreted conservatively without drawing long-term conclusions.
Key Changes
- GPT-5.5: Main leaderboard down 24.8 points, Code Execution -22 points, Material Constraints -28.3 points
- Gemini 2.5 Pro: Main leaderboard down 24.7 points, Code Execution -27.2 points, Material Constraints -21.6 points
- Gemini 3.1 Pro: Main leaderboard down 17.3 points, Code Execution -22 points, Material Constraints -11.6 points
- DeepSeek V4 Pro: Main leaderboard down 15.6 points, Code Execution -18.9 points, Material Constraints -11.6 points
- Claude Sonnet 4.6: Main leaderboard up 14.1 points, Code Execution +31.1 points, Material Constraints -6.6 points
Signals to Watch
- GLM-4.6: Today's Integrity rating is fail (based on today's Smoke data).
When reading this type of Smoke brief, the focus should be on two questions: first, whether a model exposes the same type of weakness for multiple consecutive days; second, whether the Integrity rating moves 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 real degradation, requiring follow-up runs for confirmation.
Data source: YZ Index | Run #392 | View raw data
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接