1
Gemma-4-26B-A4B with 39-69% less thinking at same accuracy
This is the fifth model in the ThinkingCap series. We reduced how much Google's gemma-4-26B-A4B-it thinks, without compromising the quality of its answers.
Building efficiency-first foundational AI. Novel AI algorithms & applications by our in-house research. Built in Europe.
Visit bottlecapai.com → On Hacker News Oct 2026 – Oct 2026
This is the fifth model in the ThinkingCap series. We reduced how much Google's gemma-4-26B-A4B-it thinks, without compromising the quality of its answers.