BottleCap AI

Maker of the token-efficient ThinkingCap reasoning models.

BottleCap AI is a Prague-based research lab building efficiency-first models, co-founded by Tomáš Mikolov, the researcher behind word2vec, alongside Jaroslav Beck and David Herel. Its ThinkingCap series fine-tunes open-weight reasoning models to cut the length of their thinking traces rather than their answer quality, with the 27B build reporting roughly half the thinking tokens of the Qwen3.6-27B checkpoint it is trained from at comparable accuracy. The weights are published on Hugging Face under Apache 2.0, which suits reasoning workloads where output-token cost dominates the bill.