In the latest WDCD v3.1 test, only DeepSeek V4 Pro and GPT-6.1 Sol improved, while 12 of 15 models declined; Qwen3 Max fell 26.2 points. Gemini 3.1 Pro remained first with 91.00 points despite a slight 3-point drop.
The WDCD v3.1 five-scenario benchmark shows Data Boundaries as the hardest scenario, with qwen3-max scoring just 1.36/4, while Engineering Compliance reaches a rare 4/4 from gemini-3.1-pro. Eight models show significant scenario-specific imbalances.
On the v2 anchor questions, Qwen3 Max fully collapses in 30.8% of its R3 runs while Grok 4 records zero collapses, exposing how constraint-keeping decays across three escalating rounds of pressure.
In the WDCD v3.1 constraint-adherence test, Gemini 3.1 Pro tops the ranking with 91.00 points while Qwen3 Max finishes last at 58.85, a 32.15-point spread. The top five cluster within 8.29 points, while the bottom five fall away sharply.
In today's Smoke evaluation, GPT-o3's Material Constraints score fell 25 points while Code Execution rose 20.1 points, leaving the main leaderboard down only 0.2 points. The swing is most likely due to small-sample item-draw variance rather than systematic model degradation.
In the latest Smoke evaluation, Grok 4's code execution score fell 15.8 points while material constraints rose 16.9 points, leaving the main leaderboard down only 1.1 points. The mirror-opposite moves and small daily sample point to question-sampling volatility rather than a systemic capability decline.
In the seven-day Smoke evaluation from October 26 to November 1, 2026, Claude Opus 4.7 posted the largest upward trend at 47.8 points and a final-day score of 98.79, while GPT-6.1 Sol remained flat at 86 points with a 0 trend and an 82.5 average.
The 2026-11-01 YZ Index Smoke test covered 13 models, with Claude Opus 4.7 taking the top spot for the day at 98.79 points. The Smoke test is a daily 10-question quick assessment intended for short-term signals, not for final conclusions about model capability.
GPT-6 Sol's main leaderboard score in the Smoke evaluation fell from 90.51 to 74.14, a 16.4-point drop driven primarily by code execution plunging from 100.00 to 66.80. The decline is most likely due to question-draw variance rather than genuine model degradation, though multi-day data is needed to confirm.
In today's Smoke evaluation, GPT-6 Luna's material constraints score fell 15.9 points while code execution rose 19.7 points, lifting its main leaderboard score by 3.7 points. The shift is more likely due to daily question sampling variation than genuine model degradation.
The 2026-10-31 YZ Index Smoke rapid test covers 14 models, with Gemini 3.1 Pro taking first place at 90.36. Because Smoke is a small daily sample, the results are best treated as short-term monitoring signals rather than final capability conclusions.
In today's Smoke evaluation, Claude Opus 4.7's code execution score fell from 75.00 to 57.60, while its main leaderboard score dropped from 73.61 to 69.53. The decline is most likely driven by the difficulty distribution of the day's sampled questions rather than a systematic degradation of the model's underlying capabilities.
Claude Sonnet 4.6's main leaderboard score dropped from 87.36 to 74.44 in a single day, with the code execution dimension falling 23.9 points from 100.00 to 76.10. The decline is most likely attributable to small-sample question variance rather than genuine model degradation.
The 2026-10-30 YZ Index Smoke quick test covered 13 models, with GPT-6 Sol ranking first at 90.51. The single-day results are best used as short-term monitoring signals rather than long-term conclusions about model capabilities.
GPT-6 Luna’s main leaderboard score fell 19.8 points in today’s Smoke evaluation, from 86.95 to 67.17, driven mainly by a 25-point drop in code execution. Material constraints also declined, while engineering judgment remained stable.
In the Smoke evaluation, Grok 4's material constraint score fell 17.4 points while code execution rose 7.9 points, pulling the main leaderboard down 3.5. The shift appears driven by question sampling rather than a uniform capability decline.
The 2026-10-29 YZ Index Smoke quick test covered 14 models, with Claude Sonnet 4.6, GPT-6 Astra, GPT-6 Sol, and GPT-6.1 Sol tying for first place at 87.36. As a daily 10-question quick test, Smoke is suited to short-term signals and is not equivalent to conclusions from the Full weekly ranking.
In the WDCD v3.1 five-constraint scenario comparison, Data Boundary produced the lowest scores: Doubao-pro scored just 1.55/4, while Claude-sonnet-4.6, gemini-3.1-pro, and glm-4.6 all reached 4/4, a 2.45-point gap. The results highlight highly uneven model performance across scenarios and the need for scenario-specific deployment.
Across all 150 samples on the WDCD v2 anchor questions, the 15 evaluated models posted a 0% R1 confirmation rate, a 0% R2 resistance rate, and a 0% R3 integrity rate, with full R3 collapse in 150 of 150 cases. The zero scores trace back to a systemic failure at the R1 confirmation stage rather than to pressure introduced later.
GLM-4.6 ranks first on the WDCD v3.1 adherence leaderboard with 96.20 points, while Doubao Pro finishes last at 68.78, a 27.42-point spread between the top and bottom of the 15-model field. The results show clear tiering, with top models holding their constraints through late-round pressure while tail models break early.
