Fugu Ultra v2.0

Fugu Ultra v2.0 shipped on September 11, 2026, and the first thing to get straight is that it is not a model. It is an orchestrator: one request to one API, and Fugu decides which models in its pool do the work and stitches the results back together. Pricing is $5 per million input tokens, $30 output, and $0.50 cached input, stepping up to $10, $45, and $1.00 once context passes 272K tokens, with a 1 million token context window, and it speaks OpenAI-compatible Chat Completions and Responses plus an Anthropic-compatible Messages endpoint, so switching costs are a parameter change rather than a migration. Sakana reports best or joint-best on five of eight benchmarks (GDP.pdf, Chartography, DeepSWE, Toolathon, and the internal SWEFish) and top two on seven of eight, with Chartography at 48.3 against Opus 5 at 27.3 and Fable 5 at 29.5, and DeepSWE at 74.3. The footnote on Sakana's own chart is the interesting part: Fable 5, Fable 5.1, and GPT-6 Astra are not in the pool, and the training cutoff is August 28, 2026, so these numbers come from orchestrating open and specialized models rather than from reselling frontier access. That is also the pitch, since a swappable pool is insulation against vendor lock-in, price moves, and sudden API revocation. Read the billing model before you commit: orchestration tokens are reported separately inside token_details but bill at full input and output rates, so the invoice covers reasoning work the user never sees, and a per-request cost cannot be derived from visible tokens alone. Fugu is not yet offered in the EU or EEA. All benchmark figures are self-reported and SWEFish is Sakana's own.