Sage-A1.5: From Long-Horizon Execution to Scientific Exploration with Capability Composition

Benchmark Model Rank Results
science-question-answering-on-frontierscience-olympiadSage-A1.5#1Accuracy: 81.0
science-question-answering-on-frontierscience-researchSage-A1.5#1Accuracy: 48.3
instruction-following-on-ifbenchSage-A1.5#2Accuracy: 81.6
physics-question-answering-on-critptSage-A1.5#2Accuracy: 20.0
code-generation-on-scicodeSage-A1.5#10Accuracy: 44.3
code-generation-on-swe-bench-proSage-A1.5#13Resolved: 61.2
deep-research-agents-on-browsecompSage-A1.5#13Accuracy: 81.7