Rick Astley - Never Gonna Give You Up (Official Video) (4K Remaster)
LR
Cognition's SWE-2 is a massive leap forward from SWE-1.7, offering higher accuracy while using far fewer steps and slashing task costs by over 80%.
Performance & Benchmarks
FrontierCode 1.1 Main: SWE-2 hits 50.0%, outperforming SWE-1.7's 42.3%.
Terminal-Bench 2.1: SWE-2 reaches 92.8%.
Comparative Standing: SWE-2 matches top models like Fable 5.1 and beats Grok 4.6 while trading blows with GPT-6 Astra at a fraction of the cost. [1, 2, 3]
Efficiency & Cost
Fewer Turns: SWE-2 (Medium mode) takes a mean of 53 steps per task on FrontierCode, compared to SWE-1.7's 127 steps (a 58% reduction).
Cost Savings: SWE-2 achieves an 81% cost reduction per task compared to its predecessor.
Comments 0
Sign in to comment
No comments yet — be the first to comment.