Google has just recently dropped a new model; Gemini 3.5 Flash. Personally, I haven’t tested it all too much, but I’ve heard that its pretty good. Google’s also teasing Gemini 3.5 Pro, which with the 3.1 lead I assume will top the benchmarks, yet not be any useful unless they tried to actually get something done.
Kimi K2.6 is my new favorite model, because its so cheap, so smart, and so conversational. I never thought I’d find a model easier to talk to than Claude; Moonshot broke those expectations.
I honestly hope that Deepmind comes and drops an absolute banger with Gemini 3.5 Pro, because I would love to actually utilize the Gemini sub I have lying around from Google Workspace.
GPT 5.5 is also a great model, yet about as conversational as a duck. In my list, I would rank the models in terms of conversibility like this: GPT < Gemini < Claude < Kimi. Kimi is also much cheaper than Claude and you can see its internal reasoning, which is a plus for debugging where the model went wrong.