Synthesis history

New items this pass are substantively empty — social media posts and article listings with no captured claims, quotes, or stances. One Instagram reel (37542) titled 'Japan can't match the US or China on AI spending — so…

Grok (xAI's AI assistant) characterized Fugu as 'a sophisticated harness/orchestration project' and 'not an open-weight model' on June 29-30[^36629][^36626], adding an external AI system's voice to the architecture deba…

Prasenjit Sarkar stated explicitly on June 28 that 'The model is no longer the product. The orchestration layer is.' [^35999], giving @jumperz's earlier reframing a second independent named voice and making it a perspec…

Rohan Paul published a substantive analysis of the Fugu technical report on June 28, adding specific architectural claims: the coordinator learns routing from data rather than rules, and Fugu-Ultra constructs distinct p…

Two meaningful framing developments emerged June 25–26. Paddo.dev published a dedicated long-form post titled 'A Multi-Agent System Sold as a Model: Sakana's Fugu,' giving the critical orchestration-vs-model distinction…

The arXiv technical report is now publicly available [^34425] and Requesty.ai published a reverse engineering of the orchestration architecture. [^34432] Informal independent testing has begun accumulating — YouTube com…

The main new development is a sustained amplification wave around the geopolitical/export controls framing, with multiple accounts posting threads explicitly foregrounding 'THE GEOPOLITICAL ANGLE' and describing Fugu as…

The export controls framing is the most significant new development: Sakana explicitly marketed Fugu Ultra as avoiding US export control risk, Chris Albon publicly questioned the claim, and enthusiast accounts amplified…

Sakana AI (Tokyo) launched Fugu and Fugu Ultra on June 22, 2026 — a multi-agent orchestration system that routes decomposed subtasks across GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro through a single OpenAI-compatible e…