Enterprise AIPaid

Fugu Max

Multi-agent orchestration that routes tasks to the cheapest capable model

sakana.ai
Overview
Fugu Max is Sakana AI's cost-performance tier of its Fugu orchestration system, launched September 11, 2026. Rather than a single monolithic model, it uses a learned router to dynamically dispatch each request across a pool of open-weight and specialized models — including NVIDIA's Nemotron family — and can recursively call instances of itself. It exposes an OpenAI-compatible API and accepts text, images, and PDFs as input.
Best for
Developers and teams wanting frontier-level task performance at significantly lower inference cost without managing their own model routing logic.
Pricing
Pay-as-you-go: $2/1M input tokens, $6/1M output tokens, $0.25/1M cached input tokens, $10/1K web search calls. Subscription plans also available via console.sakana.ai: Standard $20/mo, Pro $100/mo, Max $200/mo.

Key features

  • Learned multi-agent router dispatches each task to the leanest capable model
  • 1M-token context window with up to 128K completion tokens
  • Supports function calling, structured JSON outputs, vision, and file (PDF) inputs
  • OpenAI-compatible and Anthropic-compatible API — no migration required
  • Output pricing 40–60% below comparable frontier models (Sonnet 5, GPT 5.6 Terra, Kimi K3)

New tools drop every week.

ONE BRIEFING · EVERY FRIDAY · FREE

Free. Unsubscribe anytime.