Sakana Fugu Max and Fugu Ultra v2 Launch on OpenRouter: Pricing and a Hidden Long-Context Surcharge
Sakana AI listed two new Fugu models on OpenRouter in the early hours of September 11 — sakana/fugu-max at 05:32 UTC and sakana/fugu-ultra-v2 at 05:43 UTC — both with 1M-token context and mandatory reasoning. I pulled the live pricing from the OpenRouter models API; here are the numbers and one detail no launch post will mention.
What shipped
Fugu is not a monolithic model. Per the API metadata, it's a learned multi-agent orchestration system — a router model that dispatches sub-tasks. Both new entries carry a September 11, 2026 canonical slug (-20260911), image+file input, 1M context, and 128K max output. Reasoning is mandatory on both; Fugu Max defaults to high effort, Ultra v2 to xhigh with max supported.
What changed: the receipts
Verified from the API, per-token pricing:
- Fugu Max: $2/M input, $6/M output, $0.25/M cache read. One flat price across the full 1M context.
- Fugu Ultra v2: $5/M input, $30/M output, $0.50/M cache read — but with a pricing override: past 272,000 prompt tokens, input jumps to $10/M (+100%), output to $45/M (+50%), cache to $1/M (+100%).
Three things fall out of those numbers. First, Ultra v2 costs exactly the same as the June fugu-ultra it replaces — a silent upgrade at a flat price. Second, the 2.7x input premium for Ultra buys a 5x completion premium: heavy-output workloads pay disproportionately. Third, and this is the catch: the long-context surcharge exists only on the Ultra tier. Fugu Max keeps $2/M all the way to 1M tokens, meaning for workloads above ~272K prompt tokens the effective gap between the tiers is smaller than the headline price suggests — Max is 5x cheaper input rather than 2.5x.
Why a builder cares
Read the override field, not the price card. OpenRouter's public pricing page shows the base rate; the 272K threshold flips at your request's prompt size, so a single long-context call can silently double your bill. If you're routing long documents through Fugu Ultra v2, budget for the $10/M tier or cap input — and note the competitive frame: DeepSeek V4.1 Flash sits at $0.15/$0.60 per million on the same aggregator, 13x cheaper input than Fugu Max for those who don't need the orchestration layer.
How much does Sakana Fugu Max cost on OpenRouter?
As of September 11, 2026, Fugu Max is priced at $2 per million input tokens and $6 per million output tokens, with input cache reads at $0.25/M. Pricing is flat across its full 1M-token context window — no long-context surcharge.
Does Fugu Ultra v2 cost more above 272K tokens?
Yes. Past 272,000 prompt tokens, input pricing doubles from $5/M to $10/M, output rises from $30/M to $45/M, and cache reads double from $0.50/M to $1/M. This override applies only to the Ultra tier; Fugu Max has none.
Did Fugu Ultra v2 raise prices compared to the original Fugu Ultra?
No. The June 2026 Fugu Ultra and the new Ultra v2 carry identical base pricing ($5/$30 per million) and the same 272K-token override. The v2 upgrade ships at a flat price.