Stealth Model Beats GPT-5.6 Sol on Coding Benchmarks

A new stealth model just hit 74% on DeepSWE, beating GPT-5.6 Sol along with most other models, and reportedly also topping it on Terminal-Bench 2.1 and SWE-Bench Verified while being cheaper. @kimmonismus is betting it's another Chinese model — the benchmark leaderboard keeps flipping fast right now.