Sakana AI to pretrain its own model 'Sawara' from scratch, starting December
SakanaAILabs · x · 2026-11-13
Sakana AI explained to ITmedia why it is now pretraining its own LLM from scratch.
- The model, tentatively Sakana Sawara, is designed for autonomous AI agents and will underpin the company's products; Sawara Small pretraining starts in December, with weights planned for release in spring 2027
- Sawara Small is a pilot; a larger Sawara Large is planned, possibly exceeding several hundred billion parameters
- Sakana previously focused on post-training open models (Namazu) and model orchestration (Fugu); it argues the moment for specializing at the model layer has arrived as open-weight models improve, training knowledge spreads, and costs fall
- Agent-style RL enables end-to-end specialization on a company's own tools and workflows, sometimes beating general-purpose models
Related event: Sakana AI to Pretrain Its Own Model Sawara from Scratch(3 posts)→
More from Companies & People
- Over 1,600 Nvidia employees urge board to reinstate Jensen Huang as CEO, per WIRED — basedjensen · 2026-11-13
- ArtCraft clones Photoshop with AI as open source, has shipped 17 tools and wants to 'rebuild everything' — altryne · 2026-11-13
- AI enters ad campaigns from Coca-Cola to retail: retouching beats full generation — emmanuelvivier · 2026-11-13
- NVIDIA announces CEO transition, leaving community stunned — TinfoilTricorn · 2026-11-13
- Second-time YC founder: selling to Dropbox didn't bring answers — consistency beats intensity — garrytan · 2026-11-13
- From reform-school toilets to $5T: a timeline of Jensen Huang and NVIDIA — tibo_maker · 2026-11-13