Gemini 3.1 Pro Material Constraints Plunge 16.7 Points, Code Execution Rises 27 Points

In today's Smoke evaluation, Gemini 3.1 Pro's main leaderboard score rose from yesterday's 83.50 to 90.84, while the material constraints dimension fell from 100.00 to 83.30, a drop of 16.7 points; the code execution dimension rose from 70.00 to 97.00, an increase of 27 points.

Data facts: main leaderboard vs. dimension comparison

Yesterday, code execution was 70.00 and material constraints 100.00; today, code execution is 97.00 and material constraints 83.30. The main leaderboard consists of two core dimensions: code execution and material constraints. Today's main leaderboard score of 90.84 is 7.3 points higher than yesterday. Engineering judgment fell from 100.00 to 75.00, task expression fell from 66.70 to 51.70, and the integrity rating remained pass.

Cause analysis: question-draw fluctuation is the most likely explanation

The Smoke evaluation has only 10 questions per day, with 2 questions per dimension. A single-day loss of 16.7 points in material constraints and a single-day gain of 27 points in code execution both exceed the normal fluctuation range. Engineering judgment and task expression declined in tandem, suggesting that the day's questions may have concentrated on scenarios requiring strict material citation and task formulation. There is no multi-day consecutive data supporting a real degradation of the model, so the most likely cause is the randomness of the question draw rather than a structural decline in model capability.

Code execution at 97.00 vs. material constraints at 83.30: the same model on the same day shows a 27-point positive and 16.7-point negative contrast, pointing to question content rather than changes in model parameters.

Implications for users

Teams that prioritize code execution can gain higher confidence today, while scenarios that rely on material fidelity require additional manual checks. When enterprises select models, if they rely on Gemini 3.1 Pro daily to process documents with citation requirements, the single-day material constraints score of 83.30 suggests adding a secondary validation step. When developers build RAG pipelines, they should reserve buffer for material constraint fluctuations rather than directly using the day's output.

Strategic judgment: main leaderboard rise masks local fluctuations

The main leaderboard score of 90.84 is higher than yesterday, and the average of core capabilities is still trending upward. The single-day loss in material constraints did not change the overall ranking advantage, but it exposes dimension instability under small-sample testing. If material constraints remain below 90 in the next evaluation period, it will be necessary to determine whether it has entered a degradation range; currently there is only single-day data, which is not a reason to lower priority.

Overall, Gemini 3.1 Pro's performance today was dominated by the question draw; the fact that the main leaderboard rose 7.3 points takes precedence over the 16.7-point drop in a single dimension. Selection teams can continue to observe the three-day average of material constraints before deciding whether to adjust usage strategy.

Data source: YZ Index | Run #391 | View raw data