Sakana AI builds Sakana Sawara LLM from scratch, first model planned for spring 2027
Article timeline
The articles in this story. Times are ET.
Sakana AIOfficialSakana AI plans its own language model series, Sawara Small, for spring 2027AISakana AI says it is building a new language model series, Sakana Sawara (provisional name), from scratch, starting with pre-training. The first model, the small Sawara Small (provisional name), is planned for release in spring 2027 if development goes smoothly, as announced at a Tokyo symposium with more than 400 attendees.
hardmaruXSakana AI to build agent-native LLM Sakana Sawara from scratchAISakana AI says it is training Sakana Sawara, an agent-native large language model built from scratch, at its RSI Lab, following the Sakana AI Symposium in Tokyo. The post links to the company's English report, which says the first model, Sawara Small (provisional name), is planned for public release in spring 2027 if development stays on track.
- Sakana AIOfficialSakana AI hosts symposium on world models, Physical AI, and its Sawara language models
AISakana AI held a symposium in Tokyo on October 26 with keynote speaker Jürgen Schmidhuber and a conversation with CEO David Ha, drawing over 400 attendees. Chief Scientific Advisor Takuya Akiba announced Sakana Sawara (provisional), the company's in-house language model series, starting with Sawara Small (provisional), planned for release in spring 2027.
- Sakana AIOfficialSakana AI outlines three pillars for AI sovereignty and a new pretraining project
AIAt its symposium, Sakana AI CEO David Ha outlined three pillars: pretraining with AI agents, post-training via Sakana Namazu to adapt LLMs to Japanese and business use, and orchestration via Sakana Fugu for frontier performance without relying on a single model.