GPT-6 Luna Leads with 88.93: 2026-11-12 Smoke Quick-Test Data Brief

The 2026-11-12 YZ Index Smoke quick test covered 14 models, with GPT-6 Luna ranking first for the day at 88.93 points. Smoke is a daily 10-question quick test suited to observing short-term signals; it is not equivalent to the conclusions of the weekly Full leaderboard.

This Smoke evaluation covered only two main leaderboard dimensions: code execution and material constraint. The main leaderboard formula is 0.55 × code execution + 0.45 × material constraint. Because the daily sample size is small, single-day scores are better treated as monitoring signals rather than as long-term conclusions about model capability.

Rankings for the Day

Data Interpretation

In today's YZ Index Smoke quick test, the leading models showed clear divergence in how code execution and material constraint combined. GPT-6 Luna topped the main leaderboard at 88.93, with its 89.6 in code execution and 88.1 in material constraint forming a balanced, high-level combination; Claude Sonnet 4.6 followed closely at 88.27 on the main leaderboard, with a standout 95.2 in code execution but only 79.8 in material constraint, indicating a lopsided structure underpinned by strong code execution. GPT-6.1 Sol, Claude Opus 4.7, and GPT-6 Sol scored 84.27, 84.11, and 83.83 on the main leaderboard respectively, all maintaining code execution above 94 while all posting 71.4 in material constraint, reflecting a similar pattern of high code execution paired with low material constraint.

Several models showed notable score changes compared with the previous run under the same methodology. Claude Sonnet 4.6 rose 28.4 points on the main leaderboard, 38.9 points in code execution, and 15.6 points in material constraint; Grok 4 rose 28.1 points on the main leaderboard and 45.2 points in code execution; GPT-6 Astra rose 25 points on the main leaderboard and 45 points in code execution. These gains may stem from day-to-day question sampling variation, or they may reflect temporary differences in model performance on specific dimensions, and follow-up runs are needed to confirm stability. GPT-5.5 and GPT-o3 likewise recorded main leaderboard gains of 19.4 and 17.1 points, code execution gains of 25.2 and 23.1 points, and material constraint gains of 12.2 and 9.7 points, which also require further data validation.

On the anomaly front, Claude Opus 4.7's material constraint plunged 16.9 points, and Doubao Pro's main leaderboard score plunged 10.6 points; both may result from question sampling variation or may point to a genuine risk of degradation, and follow-up runs are needed to confirm. Qwen3 Max was not ranked because of incomplete data. Overall, the Smoke quick test remains a small-sample, single-day signal, and interpretation should remain restrained.

Key Changes

• Claude Sonnet 4.6: main leaderboard up 28.4 points, code execution +38.9 points, material constraint +15.6 points
• Grok 4: main leaderboard up 28.1 points, code execution +45.2 points, material constraint +7.2 points, integrity warn→pass
• GPT-6 Astra: main leaderboard up 25 points, code execution +45 points
• GPT-5.5: main leaderboard up 19.4 points, code execution +25.2 points, material constraint +12.2 points
• GPT-o3: main leaderboard up 17.1 points, code execution +23.1 points, material constraint +9.7 points

Signals to Watch

• Claude Opus 4.7: material constraint plunged -16.9 points
• Doubao Pro: main leaderboard plunged -10.6 points
• Qwen3 Max: data incomplete (several evaluation dimensions missing due to API failure/timeout); an automatic re-run has been scheduled, and it is not ranked in this period
• GLM-4.6: integrity rating for the day is fail (based on the day's Smoke data).

When reading this kind of Smoke brief, the focus should be on two questions: first, whether a given model repeatedly exposes the same type of weakness over multiple 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 they may be early signals of genuine degradation, and follow-up runs are needed for confirmation.

Data source: YZ Index | Run #423 | View raw data