May 2, 2026
Sliding tiles, sliding egos
Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a coding challenge
Open AI underdog shocks the giants as commenters cheer, cope, and argue rankings
TLDR: Kimi K2.6, an openly available AI model from China, won a coding contest over Claude, GPT-5.5, and Gemini. Commenters split instantly: some called it proof the old leaders are slipping, while others said one flashy test doesn’t settle who’s actually best.
The scoreboard lit a match, and the comments section poured gasoline on it. In this coding face-off, Kimi K2.6 — an openly available model from Moonshot AI — beat big-name rivals including GPT-5.5, Claude, and Gemini in a word puzzle contest built around sliding letter tiles and spotting long English words. Kimi finished first, Xiaomi’s MiMo grabbed second, and suddenly the usual “Western labs always win” storyline looked very shaky. Even funnier: Nvidia’s model face-planted before the race even started because of a coding mistake, which readers treated like the most on-brand twist imaginable.
But the real action was the community reaction. One camp was basically yelling, “Finally, a score that means something!” Commenters praised the contest for using clear win-loss results instead of vague vibes and marketing claims, with one even linking their own rankings project to say Kimi’s strong showing wasn’t a fluke. The other camp immediately hit the brakes: these tests are just one challenge, they argued, and people are going to keep posting “X beats Y” forever because AI tools are messy, inconsistent, and good at different things.
Then came the personal testimony drama. One commenter flat-out said “Kimi is really good” and claimed it beat Claude on a real compiler project, while another predicted open models could overtake cloud-only ones soon. The mood was part victory lap, part brand-war meltdown, with a side of meme energy: Kimi as the hungry upstart, Claude as the expensive favorite getting side-eyed, and everyone else suddenly checking the leaderboard twice.
Key Points
- •Kimi K2.6 from Moonshot AI won the Word Gem Puzzle challenge with 22 match points and a 7-1-0 record, ahead of MiMo V2-Pro and GPT-5.5.
- •The contest used an objective scoring system across five board sizes, rewarding words of seven letters or more and penalizing shorter words.
- •Board generation preserved more seeded words on small grids and broke up most words on 30×30 grids, which significantly affected model performance.
- •Kimi’s reported greedy sliding strategy performed best overall, especially on heavily scrambled large boards, producing the highest cumulative score of 77.
- •Nvidia’s Nemotron Super 3 did not compete because its generated code had a syntax error and failed to connect to the game server.