News for hackers. Stuff that matters.

A comment on Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a coding challenge. Parent comment

Re:Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a codi... by Galanwe on Sunday May 03, 2026 @05:13AM (#47995026)
> You would need to test 100 or more of such puzzles, widely spread across the puzzle spectrum

Would you? I am not very knowledgable on LLMs, but my understanding was that each query was essentially a stateless inference with previous input/output as context. In such a case, a single puzzle, yielding hundreds of queries, is essentially hundreds of paths dependent but individual tests?

[ Parent | on HN ]
Re:Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a codi... by tgv on Sunday May 03, 2026 @06:31AM (#47995511)
From what I understood, it's a coding challenge: the models wrote a player for that specific word game. E.g. https://github.com/rayonnant-ai/aicc/blob/main/wordgempuzzle...
[ Parent | on HN ]
Re:Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a codi... by biscuit1v9 on Sunday May 03, 2026 @05:19AM (#47995069)
Generally speaking, would you take a conclusion based only an event that happened once?
[ Parent | on HN ]