Cognition's SWE-2 Achieves 92.8 on Terminal-Bench 2.1

Cognition's SWE-2 achieves a 92.8 score on Terminal-Bench 2.1, showcasing advanced performance in software engineering tasks with 2.8T parameters and cost efficiency. The model leverages MoE architecture and proprietary training, outperforming competitors in specific benchmarks despite lower long-horizon task capabilities.

Comments

Read original article