[KongchangAI]
Benchmark

Frontier SWE-2

A frontier benchmark for software engineering that simulates multi-step tasks in real code repositories, emphasizing long-horizon engineering capability, tool calling and autonomous debugging.

Timeline (last 90 days)

Oct 29

On the Frontier SWE-2 coding test, Claude Opus 5.5 scored slightly higher than Carbon.

Expired50%

Source Articles