Use Sakana AI Models in Big-AGI.

Sakana AI in Big-AGI

Sakana AI is a Tokyo research lab, not a general-purpose serving platform: it publishes ideas first and ships models second. If you have a Sakana AI key, Big-AGI connects to it the same way it connects to any model, billed by Sakana directly to your account with no markup added. Sakana's API sends none of the CORS headers a browser needs to call a provider directly, so Direct Connection isn't available here: requests route through the Big-AGI server instead. Your key still lives in your browser rather than in a Big-AGI database, and chats are stored locally first, with sync only if you turn it on. Big-AGI also adapts its controls to Sakana's real limits: its web-search tool doesn't support the context-size levels other vendors do, so it shows up as a clean on/off toggle instead of options that wouldn't work.

A different research bet

Sakana orchestrates and retunes other labs' models instead of outscaling them: Fugu routes expert agents per prompt, and Namazu adapts open weights on Japanese and Japanese-business data. Worth knowing, given how much of the frontier optimizes for English first.

Sakana AI in Beam

Put a Sakana model into a Beam next to Claude, GPT, and Gemini, and see how a lab built on a different set of methods actually holds up on your prompt, not just on a benchmark. Fusions can then combine, cross-check, and synthesize the parallel answers instead of just picking one. Parallel runs use more tokens than a single chat.