Hands-on with M3.1 Flash in MiniMax Code: real bug found, self-written tests 12/12 green
Testing M3.1 Flash in the redesigned MiniMax Code, the author saw it find the real bug instead of patching around it, refuse to pad weak results, write its own tests to 12/12 green, and mark things UNVERIFIED.
MiniMax Code got a full redesign, but M3.1-flash really stood out to me it makes better calls while working > it found the real bug instead of patching around it > refused to pad weak results > wrote its own tests and got to 12/12 green > and said UNVERIFIED when it couldn’t confirm something so i pointed it at my biggest problem: distribution now it finds 5 threads worth replying to then drafts a help-first reply for each, and gives me fresh drafts every morning at 7
This post quotes another post: x.com/minimaxagent/status/2104931188435771868