Sakana AI Launches Fugu Max and Fugu Ultra v2, Multi-Agent Orchestrators Up to 60% Cheaper Than Sonnet 5

My take: This says less about the model itself and more about how the market is changing: Sakana AI did not launch a model that competes head-on with GPT-6 or Claude Fable, but one that orchestrates them. Fugu Max and Fugu Ultra v2 are systems that route complex tasks across teams of smaller, cheaper models and combine the results at the end. The entry price is competitive: Fugu Max lists at $2 per million input tokens, which Sakana says is 40 to 60 percent cheaper than Sonnet 5 or Kimi K3. Fugu Ultra v2, the quality-first version, reports 74.3 on DeepSWE, a real-world software engineering benchmark.

Worth noting: those numbers come from Sakana AI itself, not an independent evaluator. The direction is clear, but it is worth waiting for external replications before treating those figures as settled.

For anyone building with AI today, the pattern that matters is this: you no longer need to use the most powerful model for every task. Orchestration systems like Fugu allow you to combine models based on the cost and capability each subtask requires, which can reduce spend without sacrificing results. Does your current architecture have that level of granularity, or are you still paying frontier prices for tasks a lighter model could handle just as well?