ResearchBenchmarksCheckpoint Selection on REPO-PROMPTS, MS-COCO, and PartiPromptsCheckpoint Selection on REPO-PROMPTS, MS-COCO, and PartiPromptsFollow4.42Intent ScoreCheck-in2.04882.66443.283.8956Mar 24, 2026Evaluation resultsMethodMethodLinksIntent ScoreHuman PreferenceCheck-in2026.034.42100Stylus2026.033.6882.1Reranker2026.033.2185.7Random2026.032.1489.1