Ornith-1.5: From Self-Scaffolding to Self-Improvement

Benchmark Model Rank Results
code-generation-on-deepsweOrnith-1.5-35B-A3B (Claude Code)–Pass@1: 0.22
code-generation-on-nl2repoOrnith-1.5-35B-A3B–Accuracy: 46.2
code-generation-on-swe-bench-multilingualOrnith-1.5-35B-A3B (OpenHands)–Accuracy: 71.4
code-generation-on-swe-bench-proOrnith-1.5-35B-A3B–Resolved: 59.6
code-generation-on-swe-bench-verifiedOrnith-1.5-35B-A3B (OpenHands)–Accuracy: 79.0
code-generation-on-terminal-bench-2-1Ornith-1.5-35B-A3B (Claude Code)–Accuracy: 68.5
deep-research-agents-on-browsecompOrnith-1.5-35B-A3B–Accuracy: 67.6
multiple-choice-question-answering-mcqa-on-gpqa-diamondOrnith-1.5-35B-A3B–Accuracy: 89.2
question-answering-on-humanity-s-last-exam-hleOrnith-1.5-35B-A3B–Accuracy: 33.4