Cognition's SWE-2 achieves 92.8 on Terminal-Bench 2.1
Cognition's new software engineering model, SWE-2, scored 92.8 on the Terminal-Bench 2.1 benchmark, a test of language model performance on terminal tasks. The result, posted on Hacker News, shows the model's strong ability to understand and generate terminal commands, indicating progress in AI-driven programming assistance. This achievement places SWE-2 among the top-performing models in the benchmark, suggesting its potential for real-world developer tools.