Claude Opus 4.7 Leads with 47.8 Trend Score as GPT-6.1 Sol Holds at 86 for 7 Days

In the seven-day Smoke evaluation from 2026-10-26 to 2026-11-01, Claude Opus 4.7 recorded a trend score of 47.8 and a final-day score of 98.79, making it the model with the largest increase.

Data Facts for Steadily Rising Models

GPT-6 Astra scored 83.49 on the first day and 97.25 on the final day, with a trend of 13.8 and an average of 86.1. Claude Sonnet 4.6 scored 75.73 on the first day and 86.38 on the final day, with a trend of 10.7 and an average of 78.9. GPT-6 Luna scored 62.75 on the first day and 90.93 on the final day, with a trend of 28.2 and an average of 77.8. GPT-6 Sol scored 62.62 on the first day and 85.62 on the final day, with a trend of 23 and an average of 78. Grok 4 scored 58.57 on the first day and 82.75 on the final day, with a trend of 24.2 and an average of 71.8. Doubao Pro scored 47.92 on the first day and 59.9 on the final day, with a trend of 12 and an average of 66.6. Gemini 3.1 Pro scored 43.09 on the first day and 61 on the final day, with a trend of 17.9 and an average of 67.1. DeepSeek V4 Pro scored 41.99 on the first day and 50.05 on the final day, with a trend of 8.1 and an average of 57.1. GPT-o3 scored 35.46 on the first day and 77.49 on the final day, with a trend of 42 and an average of 67.1. GPT-5.5 scored 32.15 on the first day and 69.06 on the final day, with a trend of 36.9 and an average of 67.6. GLM-4.6 scored 30.39 on the first day and 61.25 on the final day, with a trend of 30.9 and an average of 46.8.

Flat Models and Zero-Score Models

GPT-6.1 Sol scored 86.05 on the first day and 86 on the final day, with a trend of 0 and an average of 82.5. Gemini 2.5 Pro scored 49.95 on the first day and 50 on the final day, with a trend of 0 and an average of 58.3. Qwen3 Max scored 0 on the first day and 0 on the final day, with a trend of 0 and an average of 0.

Analysis of the Causes of High-Volatility Models

Claude Opus 4.7 had volatility of 47.8, GPT-5.5 volatility of 47, Gemini 3.1 Pro volatility of 47.3, GPT-o3 volatility of 43, Doubao Pro volatility of 43.1, and DeepSeek V4 Pro volatility of 42.9. These values come directly from standard deviation calculations of scores on the daily 10-question quick test; the max(0,100-stddev×2) formula indicates low consistency. GPT-6 Luna had volatility of 28.2, GPT-6 Sol volatility of 28.6, Grok 4 volatility of 25.3, and GLM-4.6 volatility of 30.9, likewise reflecting score fluctuations when answering similar questions multiple times.

Signals of Changes in Integrity Ratings

DeepSeek V4 Pro's integrity rating showed the sequence pass→pass→pass→pass→warn→pass→pass. GLM-4.6's integrity rating showed the sequence fail→pass→pass→pass→pass→pass→pass. GPT-6.1 Sol, GPT-6 Astra, Claude Sonnet 4.6, GPT-6 Luna, GPT-6 Sol, Grok 4, Claude Opus 4.7, Gemini 2.5 Pro, Doubao Pro, Gemini 3.1 Pro, GPT-o3, GPT-5.5, and Qwen3 Max were pass throughout.

Implications for Users

Teams focused on code execution can prioritize GPT-6.1 Sol, whose average is 82.5 and trend is 0, showing stable execution-dimension performance over the seven days. GPT-6 Astra is suitable for scenarios sensitive to material constraints; its final-day score of 97.25 and trend of 13.8 indicate continued improvement in its handling of this area. Developers who depend on consistent output should avoid Claude Opus 4.7, because its volatility of 47.8 means scores vary widely when the same type of prompt is run multiple times.

Strategic Judgment

GPT-6.1 Sol held at 86 points with an average of 82.5, maintaining the highest stable range among all models and making it suitable as a baseline model. Claude Opus 4.7's 47.8 trend score coexists with 47.8 volatility; the rise was driven by a single-day high score rather than a consistent overall improvement. After GLM-4.6 shifted from fail to pass, its trend was 30.9, with the improvement in integrity rating accompanied by a score increase. Qwen3 Max had an average of 0 and a trend of 0, with no usable signal in this period's data. After a warn appeared for DeepSeek V4 Pro, its trend was only 8.1; integrity fluctuations did not translate into sustained high scores. The next period should verify whether the trends of GPT-6 Astra and GPT-o3 remain above 13.8 and 42, respectively.

Data source: YZ Index | Run #405 | View raw data