DeepSeek V4 Pro Material Constraint Score Drops 19.5 Points in a Single Day, Main Ranking Slips 7.6 Points

DeepSeek V4 Pro's material constraint score in today's Smoke evaluation fell from 89.50 yesterday to 70.00, while its overall main ranking dropped from 94.07 to 86.50.

Score Facts and Dimension Breakdown

Code execution rose from 97.80 to 100.00, engineering judgment held steady at 100.00, task expression dipped slightly from 95.80 to 95.00, and the integrity rating remained at pass. The -19.5-point drop in material constraint directly pulled the main ranking down by 7.6 points.

Cause Analysis: Sampling Fluctuation or Model Degradation

The Smoke evaluation uses only 2 questions per dimension per day, so the single-day plunge in material constraint most likely stems from an increase in question difficulty. Code execution simultaneously rising to a perfect score suggests that the model's overall capability has not undergone systematic degradation; the fluctuation is closer to single-dimension sampling variance.

Material constraint 70.00 vs. 89.50 yesterday, a difference of 19.5 points; code execution 100.00 vs. 97.80 yesterday, a difference of +2.2 points.

If this were genuine degradation, code execution and engineering judgment would typically be affected in tandem. However, the current data shows only material constraint as anomalous, pointing to a sudden increase in the fidelity requirements of the test questions.

Specific Implications for Users

Scenarios heavily reliant on material constraint—such as legal contract extraction or academic citation verification—require additional manual review, as a single-day score of 70.00 could cause output to deviate from source materials. The perfect score in code execution remains meaningful reference value for teams focused on code debugging.

• Developers who depend on material constraint should add multiple rounds of validation before critical tasks.
• Users with stable code execution results can continue using the model for programming-related work.

Strategic Assessment

A single-day 19.5-point swing in material constraint exceeds the normal sampling range, and the next evaluation cycle should prioritize verifying whether this dimension remains persistently low. Current data does not support a conclusion of overall model degradation, but the instability in material constraint warrants priority attention from model selection teams.

Data source: YZ Index | Run #270 | View raw data