Fugu Ultra v2.0

Sakana AI released Fugu Max and Fugu Ultra v2 on September 11, 2026, describing Fugu as an orchestrator that takes a single API call and internally dispatches to multiple open-weight models in its pool rather than acting as a conventional large language model. On the Chartography visual-reasoning and data-interpretatio The briefing frames Sakana AI's Fugu as a model-agnostic orchestration product, decoupling frontier-level output from any single proprietary model by routing requests through a swappable pool of open-weight and specialized models. This positioning is presented as a structural shift in AI economics: when the orchestrati

The Pasquale Pillitteri analysis confirms the September 11, 2026 dual release of Fugu Ultra v2.0 (capability ceiling) and Fugu Max (budget orchestrator) by Sakana AI, the Tokyo startup founded by Llion Jones and David Ha. It restates the headline result: Fugu Ultra v2 beats GPT-6 Astra and Claude Fable 5.1 on several b Beyond launch facts, the article adds regional availability context, noting that Fugu Ultra v2 remains locked out of the EU even as it is available in the US and UK, and references NVIDIA Nemotron's role in Fugu Max's orchestration pool. The piece frames the release as a shift toward routing through teams of models rat

Sakana AI launched Fugu Max v1.0 and Fugu Ultra v2.0 on September 11, 2026 as API-compatible multi-agent orchestration products rather than monolithic models. The architecture rests on learned model coordination described in two ICLR 2026 papers: TRINITY, which uses an evolved LLM coordinator, and Conductor, which appl Fugu Ultra v2 reaches best or joint-best results on five of eight benchmarks, including GDP.pdf, Chartography, DeepSWE, Toolathon, and SWEFish. Pricing is $5 per million input tokens, $30 per million output tokens, and $0.50 per million cached input tokens, with a premium tier of $10 / $45 / $1.00 for contexts above 27

Sakana AI released Fugu Ultra v2.0 on September 11, 2026 as the capability-first tier of its Fugu family, served through Sakana's OpenAI-compatible API under the model id fugu-ultra-v2.0. The model is not a single trained network but a Multi-Agent System that dynamically coordinates a pool of models behind one endpoint Fugu Ultra v2.0's two published benchmark figures are 48.3 on Chartography, a visual-reasoning and data-interpretation benchmark, compared with Opus 5 at 27.3 and Fable 5 at 29.5, and 74.3 on DeepSWE, a real-world software engineering benchmark; Sakana also reports that the model tops SWEFish. Pricing is set at $5 per