Gemma-4-26B-A4B with 39-69% less thinking at same accuracy
The developers of the 'ThinkingCap' series have released a new model optimized for Google's Gemma-4-26B, significantly reducing the number of tokens used for 'thinking' without sacrificing accuracy. This efficiency improvement aims to lower inference costs and increase speed for enterprise users.
Why it matters
Optimizing large language models for efficiency is critical for making advanced AI more accessible and cost-effective for large-scale enterprise deployment.
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.
ThinkingCap PRO: want even better efficiency? The PRO version of this model uses up to 87.6% fewer thinking tokens, 62% on average. Faster answers and lower inference costs at scale. Contact us at enterprise@bottlecapai.com .
We have published four token-efficient models so far: ThinkingCap-Qwen3.6-27B , ThinkingCap-Qwen3.8-27B , ThinkingCap-NVIDIA-Nemotron-3.5-Lightning-30B-A3B and ThinkingCap: Qwen3.8-Flash-Next .
gemma-4-26B-A4B is Google's sparse mixture of experts — 26B parameters with about 4B active per token. It is the third model family we have applied this work to, after Qwen and NVIDIA. On this one we put extra effort into keeping the accuracy intact: thinking comes down 38.9% on average at a cost of only 0.15pp of accuracy, which is smaller than the run-to-run spread on most of the suite.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in