Hi, I'am seeing intermittent hangs with

kimi-k2-thinking

during SWE-bench evaluation on remote runtime. The agent appears to finish, but the

/api/conversations/{id}/run

endpoint never returns, causing 30-minute timeouts. Switching to

kimi-k2-0905

resolves the issue completely. Is this a known bug? Thanks for the help!

Which inference engine are you using, or which API provider?