Sakana AI to pretrain first domestic model Sakana Sawara for defense and cybersecurity by 2027
tkasasagi · x · 2026-10-26
- Sakana AI announced it will build its own generative AI model from scratch, including pretraining. The model, tentatively named Sakana Sawara, enters development in November with a small model slated for spring 2027; the large model is expected to have hundreds of billions of parameters or more, likely smaller than leading US/China frontier models.
- The move addresses sovereign-AI concerns: its existing Sakana Fugu stack depends partly on foreign models. Sawara targets defense and cybersecurity use cases — Sakana signed an AI intelligence-analysis pilot contract with Japan's Ministry of Defense in July.
- Context: per Artificial Analysis, Meta's Muse-Wonder and Moonshot AI's Kimi K3.1 Pro both exceed GPT-6; DeepSeek and Xiaomi are pushing cheap high-performance models, intensifying frontier-scale competition. Sakana plans AI-assisted development to cut cost and time.
More from Companies & People
- Google DeepMind's Frontier Strategy & Governance team is hiring in London — AllanDafoe · 2026-10-26
- X's For You checks only 2,800 posts per open — and a new fair-share filter stops fresh posts being cut first — faraz0x · 2026-10-26
- Leadership tips from a former WoW mythic raid leader: reduce cognitive burden — Linahuaa · 2026-10-26
- Revolut CEO: China's 9/9/6 work culture makes it unbeatable — only 2 of 140 startups survived — davidsenra · 2026-10-26
- Springer Nature Partners With Google DeepMind on AI Peer Review Assistant — GoogleDeepMind · 2026-10-26
- Nearly 300 indie hackers turn out in Dubai; over half have earned at least $1 — marclou · 2026-10-26