Introducing K2 Horizon: Frontier Performance, Radically Open

Benchmark Model Rank Results
mathematical-reasoning-on-hmmt-feb-2026K2-Horizon-7B#8Accuracy: 73.3
agents-on-toolathlon-verified-108-tasksK2-Horizon-375B-A23B#10Pass@1: 0.653
agents-on-automationbench-publicK2-Horizon-375B-A23B#12Success Rate: 25.3
code-generation-on-scicodeK2-Horizon-7B#23Accuracy: 31.6
deep-research-agents-on-browsecompK2-Horizon-375B-A23B#26Accuracy: 72.8
question-answering-on-humanity-s-last-exam-hleK2-Horizon-7B#31Accuracy: 18.6
code-generation-on-swe-bench-verifiedK2-Horizon-7B#39Accuracy: 70.6
code-generation-on-terminal-bench-2-1K2-Horizon-7B#44Accuracy: 39.1