Rufus-Air: An Open LLM Post-Training Recipe

Benchmark Model Rank Results
agents-on-bench-airlineRufus-Air–Accuracy: 84.0
agents-on-bench-retailRufus-Air–Accuracy: 85.3
agents-on-bench-telecomRufus-Air–Accuracy: 93.0
code-generation-on-livecodebench-v6Rufus-Air–Pass@1: 0.764
code-generation-on-swe-bench-verifiedRufus-Air–Accuracy: 65.6
code-generation-on-terminal-bench-2-1Rufus-Air–Accuracy: 42.7
instruction-following-on-ifbenchRufus-Air–Accuracy: 76.9
instruction-following-on-ifevalRufus-Air–Prompt-level strict-accuracy: 95.4
instruction-following-on-multichallengeRufus-Air–Accuracy: 65.8
mathematical-reasoning-on-aime-2026Rufus-Air–Accuracy: 86.7
multiple-choice-question-answering-mcqa-on-gpqa-diamondRufus-Air–Accuracy: 75.6
science-question-answering-on-frontierscience-olympiadRufus-Air–Accuracy: 54.6
science-question-answering-on-frontierscience-researchRufus-Air–Accuracy: 35.1