AI analyst Lisan al Gaib on why the consensus that Kimi K3 is frontier is wrong: "There's always this debate of how good is the model and how far is it behind the frontier. When I say Kimi K3 hasn't caught up to US models, I don't mean this in a bad way, but from a forecasting perspective, I'm very stats driven. You also have to differentiate between different domains." "When you look at coding, I think Kimi has basically caught up to within a few months, maybe three or four months behind the frontier. But if you look at overall capabilities, then it's more like five months, and maybe longer if you include Mythos preview." "Many people don't realize how large or how tiny frontier models from OpenAI or Anthropic are. Another aspect is reasoning efficiency. When we look at these benchmarks, Kimi is still one or two generations behind in reasoning efficiency. But in terms of potential, I have to agree that Kimi can basically reach Fable level and GPT-5.6 Sol level, just with more post-training." @scaling01