Announcing Bonsai 27B: The First 27B-Class Model to Run on a Phone

Benchmark Model Rank Results
mathematical-reasoning-on-math-500Ternary Bonsai 27B#2Accuracy: 99.20
instruction-following-on-ifevalTernary Bonsai 27B#12Prompt-level strict-accuracy: 85.03
agents-on-benchTernary Bonsai 27B#19Accuracy: 73.61
mathematical-reasoning-on-aime-2025Ternary Bonsai 27B#19Accuracy: 90.84
multi-task-language-understanding-on-mmlu-reduxTernary Bonsai 27B#20Accuracy: 88.05
mathematical-reasoning-on-aime-2026Ternary Bonsai 27B#22Accuracy: 87.50
multimodal-reasoning-on-mmmu-proTernary Bonsai 27B#22Accuracy: 68.96
instruction-following-on-ifbenchTernary Bonsai 27B#29Accuracy: 58.50