The new chemical reasoning model “ether0” has been released

FutureHouse has unveiled an efficient inference model named ether0, a 24-billion-parameter model specifically designed for scientific reasoning in the field of chemistry. Backed by former Google CEO Eric Schmidt, this startup has enhanced the model’s chemical capabilities through post-training, while requiring significantly less data compared to traditional specialized models. The research team employed various techniques, including reinforcement learning and inference behavior distillation, enabling ether0 to outperform GPT-4.1 and DeepSeek-R1 across multiple chemistry tasks—achieving twice the accuracy of its competitors in some cases. Additionally, ether0 demonstrates strong performance in understanding natural language queries and generating molecular structures, making it particularly well-suited for drug-like molecule design.