Fugu Playground

Chat with Fugu Ultra and Sakana Namazu right here, using your own OpenRouter API key. This is a bring-your-own-key tool: your browser talks to OpenRouter directly, your key lives only in this tab’s sessionStorage, and this site’s server never sees your key, your prompts, or the responses.

Before you start: create a dedicated key with a credit limit in your OpenRouter dashboard. Fugu Ultra reasons on every request (it can’t be turned off) and bills accordingly — treat every message as a paid API call, because it is one.

You pay for every message. This playground uses your own OpenRouter API key and bills your OpenRouter account. As of August 2026, Fugu Ultra costs $5 per 1M input tokens and $30 per 1M output tokens (more above 272K context), and its reasoning cannot be turned off — a single long answer can cost anywhere from a few cents to over a dollar. We recommend a dedicated key with a credit limit. This site is unofficial and never sees your key or your prompts.

No key set. Your key is kept in this browser tab's sessionStorage only — it is never sent to fugu-ai.com, never logged, and is gone when you close the tab.

Namazu is a different Sakana model (Japanese-specialized) — cheaper for pipeline testing: $0.95/M input, $4.00/M output as of August 2026.

Fugu Ultra always reasons (it cannot be disabled). Higher effort = better answers, more tokens, higher cost.

Session cost: $0.0000

“Stop” only stops rendering in your browser. Sakana streams cannot be aborted server-side, so the full response may still be generated — and billed — on your OpenRouter account.

Unofficial tool. Requests go from your browser straight to openrouter.ai; per OpenRouter's listing, the provider (Sakana AI) retains prompts but does not train on them — see the model page for its current data policy. Costs shown are estimates based on OpenRouter's published rates (August 2026) or OpenRouter's reported usage; your OpenRouter dashboard is the source of truth.

Good to know

  • Why OpenRouter? It explicitly allows cross-origin requests, which is what makes a purely client-side, server-never-involved playground possible. It lists both Sakana models at the same per-token rates as the direct API (as of August 2026), and its keys support per-key credit limits — ideal for experiments.
  • Why is there no temperature slider? Fugu ignores sampling parameters like temperature and top_p, so we don’t send any.
  • Why does Stop keep billing? Sakana streams can’t be aborted server-side. Stop halts rendering in your browser, but the provider may finish generating — and billing — the full response.
  • What’s the cost figure under each reply? OpenRouter’s reported usage for that exact request when available, otherwise an estimate from the published per-token rates. Your OpenRouter dashboard is always the source of truth.

Planning a bigger workload? Estimate it first with the Fugu Cost Calculator.