Summary
On September 11, 2026, Sakana AI released Fugu Ultra v2 (capability-first) alongside Fugu Max (cost-first). Fugu is a multi-agent system delivered as a single OpenAI-compatible model that dynamically routes and orchestrates a pool of underlying models. Fugu Ultra v2 has a 1M-token context and leads five of eight benchmarks in Sakana's own evaluation, including 48.3 on Chartography versus Claude Opus 5 at 27.3.
What changed
Sakana AI shipped Fugu Ultra v2 and Fugu Max. Both expose a single OpenAI-compatible endpoint that dynamically coordinates a pool of models rather than serving one trained network. Fugu Ultra v2 accepts text, images, and files including PDFs, has a 1M-token context and up to 128K completion tokens, and prices at $5.00/$30.00 per million input/output tokens (cache read $0.50/M; web search $10 per 1K calls). It is best or joint-best on five of eight benchmarks in Sakana's evaluation (48.3 Chartography vs Claude Opus 5 at 27.3; 74.3 DeepSWE).
Why it matters
Fugu packages model orchestration, usually a builder's job, into a single callable model, competing with routing layers like Vercel AI Gateway and OpenRouter and with frontier single models. If the orchestration genuinely beats individual models on price-performance, it reframes 'pick a model' as 'pick an orchestrator.'
Evidence excerpt
Fugu is a Multi-Agent System, Delivered as One Model, reaching results by dynamically coordinating and orchestrating a diverse pool of powerful models behind one OpenAI-compatible endpoint. Fugu Ultra v2 is best or joint-best on five of eight benchmarks, including 48.3 on Chartography against Claude Opus 5 at 27.3, and 74.3 on DeepSWE.