The Inference Report

The SWE-rebench rankings show minimal movement at the top tier, with gpt-5.5-2026-04-23-xhigh holding position one at 62.7% and Junie maintaining second place at 61.6%, but the middle and lower portions of the leaderboard reveal substantial volatility. Gemini 3.1 Pro Preview dropped from 57.2% to 51.1%, falling from fifth to eleventh place, while Gemini 3.5 Flash fell from 55.3% to 49.5%, sliding from eighth to thirteenth. Kimi K2.6 declined from 53.9% to 46.5%, surrendering its tenth-place position to land at fifteenth. Conversely, GLM-4.7 improved from 42.1% to 38.2%, though this represents a methodological concern: the Artificial Analysis benchmark shows GLM-4.7 at 42.1% while SWE-rebench reports 38.2%, raising questions about whether these measure comparable problem-solving capabilities or whether the SWE-rebench evaluation may have shifted its difficulty calibration. Claude Sonnet 4.6 moved upward from eighteenth to tenth on Artificial Analysis, gaining 0.4 points to reach 51.7%, suggesting incremental refinement rather than breakthrough performance. The divergence between the two benchmarks across the same models underscores that coding ability assessments depend heavily on test selection and evaluation methodology, making absolute rankings less informative than the specific gaps they reveal about model strengths on particular problem classes.

Cole Brennan