The most direct finding from today's Smoke lightweight benchmark is that code execution ability no longer differentiates the top seven models, as all scored 100, making rankings entirely determined by material constraint scores.
Smoke's lightweight evaluation completed 10-question quick test on 11 mainstream models. Claude Sonnet 4.6 scored 99.78 points, while the GPT series collectively stuck at 74 points, highlighting the polarization in AI capabilities.
In the latest WDCD cycle, all 11 evaluated models show improvement in compliance ability, with the top five all rising and none declining. Gemini 3.1 Pro leaps into the top three with a +14.2 point gain, signaling a major shift in the competitive landscape.
The WDCD compliance test evaluates model stability under real enterprise constraints through three rounds of dialogue. The resource limitation scenario scored the lowest overall, becoming a common "stumbling block" for all 11 models.
Eleven mainstream models showed a clear degradation trajectory in the three-round WDCD test: nearly all confirmed constraints in R1, maintained 93% resistance after R2 interference, but average integrity rate dropped to only 30.5% in R3, with 200 tests directly hitting zero.
The first public ranking of the WDCD compliance test shatters the myth that bigger parameters mean greater reliability. Qwen3 Max leads with 70.83 points, while Grok4 finishes last with 51.67 points; the average crash rate in Phase R3 reaches 60.6%, proving that most models are still highly prone to violating constraints under real enterprise conditions.
This week's Smoke rapid tests over 7 consecutive days reveal DeepSeek V4 Pro's steep decline from 97.08 to 66.88, averaging 79.8 with high volatility. In contrast, GPT-5.5 and Claude Sonnet 4.6 show steady rebounds, with GPT-5.5 rising 11.5 points.
In today's Smoke quick test, ERNIE Bot 4.5's main leaderboard score fell from 74 to 62.96, a drop of 11 points, with code execution collapsing from 100 to 50 points, while material constraints only edged up 4.5 points.
Ernie Bot 4.5's execution score dropped sharply from 100 to 50, causing its main leaderboard score to plummet 11 points to 62.96. This is not a minor fluctuation but a clear collapse in core capabilities.
DeepSeek V4 Pro delivered extremely polarized results in today's Smoke evaluation. The main index jumped from 39.26 to 87.99, a gain of 48.7 points; the code execution dimension soared from 20.00 to 100.00, while material constraints saw a modest increase of 10.5 points. However, engineering judgment (side index, AI-assisted evaluation) plummeted from 38.40 to 10.00, a drop of 28.4 points.
In the latest Smoke Lite benchmark results, Claude Sonnet 4.6 leads the main leaderboard with 91.77 points, achieving a perfect 100 in code execution and 81.7 in material constraints. GPT-o3 scores only 50 in execution, ranking last with 62.83 points.
Doubao Pro's main score in today's Smoke evaluation dropped from 81.33 to 40.12, a decline of 41.2 points, primarily due to the code execution dimension collapsing from a perfect 100 to 20, losing 80 points in a single day.
Gemini 3.1 Pro's code execution score plummeted from 100.00 to 20.00 in today's Smoke evaluation, causing a 33.5-point drop in the main rankings. This is not a minor fluctuation but a near-total failure of a core capability in a single day's test.
In the Smoke evaluation released at 3 AM today, all 11 mainstream models experienced a collective crash on the main leaderboard, with an average drop of 42 points. Gemini 3.1 Pro topped the list with 40.48 points, but this score itself dropped 33.5 points from yesterday, with only 20 points remaining in the execution dimension and 65.5 points in the constraint dimension.
The most significant finding in this WDCD cycle is that Qwen3 Max topped the chart with 72.50 points, a 15-point jump from Run #125, while Claude models saw notable declines, with Opus 4.7 dropping 7.5 points and Sonnet 4.6 falling to second place but now trailing the leader by 7.5 points.
The WDCD five-scenario review reveals that business rules are the weakest area for all 11 models, with an average score of only 2.05, while security compliance shows the largest differentiation with a 2-point gap between highest and lowest scores.
The WDCD three-round test reveals models' true reliability under pressure: the R1 confirmation rate of 93% plummets to 26.4% integrity in R3, with most models collapsing after initially complying.
Qwen3 Max scored 72.50 points, leading the WDCD compliance test by 7.5 points over second-place Claude Sonnet 4.6, while ERNIE Bot 4.5 scored 45 points as the only model below 50, and the 60.9% R3 breakdown rate exposed the industry's weakness under adversarial pressure.
Claude Sonnet 4.6 showed significant anomalies in today's Smoke evaluation, with the material constraint dimension dropping directly from 74.50 to 59.50, a single-day decline of 15 points, and the main ranking dropping 6.8 points to 81.78. This magnitude exceeds the normal random fluctuation range of the daily 10-question quick test.
Claude Opus 4.7's material constraint dimension plummeted 15 points in today's Smoke evaluation using only 10 questions, from 74.50 to 59.50, while the overall leaderboard fell 6.8 points to 81.78.
