The Inference Report

The SWE-rebench leaderboard shows no movement in the top positions, with OpenAI's gpt-5.5-2026-04-23-xhighModel holding rank #1 at 62.7% ± 0.91%, followed by JunieJunieAgent at 61.6% ± 0.64% and OpenAICodexAgent at 60.4% ± 1.37%, the same three-model ordering from the previous cycle. Artificial Analysis, by contrast, registered a single meaningful shift: Claude Opus 5 debuted at #1 with 60.7, displacing Claude Fable 5 from the top spot to #2 at 59.9, a 0.8-point gain that exceeds typical measurement noise given the benchmark's established variance patterns. The SWE-rebench confidence intervals (ranging from 0.45% to 1.98%) suggest sufficient precision to detect real movement, yet the absence of any ranking changes across 24 models indicates either genuine stability in relative performance or that improvements, if present, fall within the bounds of statistical uncertainty. Artificial Analysis's deeper leaderboard shows broader reshuffling throughout the 400-entry list, though the top tier remains concentrated among OpenAI, Anthropic, and emerging Chinese models like Kimi and GLM variants. Without prior Artificial Analysis data to establish trend patterns, it remains unclear whether the Claude Opus 5 promotion reflects sustained capability gains or transient evaluation variance; the SWE-rebench's frozen ranking, however, suggests that whatever changes occurred in the underlying systems did not cross the performance thresholds needed to alter the established hierarchy.

Cole Brennan