Fugu Subscription Plans: Standard vs Pro vs Max

Fugu’s subscription plans are the simple way in: a flat monthly fee, both Fugu and Fugu Ultra included, no token math required. But the three tiers are defined in an unusual way — relative to each other rather than in absolute numbers — and there’s a documented catch about priority that matters if you care about speed. Here’s the full picture, verified against Sakana’s official pricing page as of August 2026.

The three plans

Plan	Price	Usage allowance	Models included
Standard	$20/month	Baseline allowance	Fugu + Fugu Ultra
Pro	$100/month	10× Standard	Fugu + Fugu Ultra
Max	$200/month	20× Standard	Fugu + Fugu Ultra

Every tier includes both models — you’re never paying extra to unlock Fugu Ultra. What you’re buying with the higher tiers is quantity.

What “baseline, 10×, 20×” actually means

Notice what’s missing from that table: a number. Sakana does not publish the Standard plan’s allowance in tokens, requests, or dollars — it’s simply “baseline,” with Pro and Max defined as ten and twenty times that. This relative definition gives Sakana room to tune the baseline over time, and it means nobody can tell you in advance exactly how much usage $20 buys.

What you can reason about is the exchange rate between tiers. Pro costs 5× Standard’s price for 10× its allowance — per unit of usage, half the price. Max costs 10× Standard’s price for 20× the allowance — the same per-unit rate as Pro, just more of it. In per-usage terms, Standard is the most expensive tier and exists as the low-commitment entry point; Pro and Max are the same deal at different scales.

The priority catch

One line from Sakana’s pricing page deserves more attention than it gets: “Consumption-based tokens are served at higher priority than monthly plan tokens.” Pay-as-you-go traffic goes first; subscription traffic yields. Sakana frames pay-as-you-go as the option for “heavy production workloads that need maximum reliability,” and the priority rule is why.

Practical translation: a subscription is a fit for interactive, human-paced work, where you won’t notice yielding to production traffic. It’s the wrong fit for anything where latency or throughput guarantees matter — that’s pay-as-you-go territory, priced per token on the main pricing page.

Subscription vs. pay-as-you-go: the real decision

The two payment modes differ on more than price. Pay-as-you-go runs on prepaid credits that expire six months after purchase and are non-refundable; a subscription is a recurring fee with no credit inventory to manage. Pay-as-you-go costs scale with usage — including orchestration tokens, the internal usage Fugu Ultra bills at full rates; a subscription’s flat fee absorbs that variability, which is genuinely valuable given that per-request costs on Fugu Ultra can’t be hard-capped.

A useful mental model: subscriptions transfer cost risk to a flat fee, pay-as-you-go transfers scheduling priority to you. Pick by which risk you’d rather hold.

Who should pick what

• Curious individual, first month with Fugu: Standard at $20. Both models, minimal commitment, and you’ll learn your real usage pattern before spending more.
• Daily driver for coding: Standard until you hit its ceiling, then Pro. Coding-agent use (Codex or Claude Code with Fugu behind them) is exactly the human-paced, interactive load subscriptions suit.
• Power user running long Ultra sessions: Max. Twenty times the baseline at Pro’s per-unit rate — the tier for people whose work lives inside these models.
• Production API traffic, latency-sensitive pipelines: skip subscriptions. Pay-as-you-go is served first and scales without an allowance ceiling — model your spend with the cost calculator before committing.
• Teams doing both: the modes aren’t mutually exclusive — interactive work on plans, production on credits, and the priority rule works in your production traffic’s favor.

What every tier includes — and what none does

All three plans share the same catalog: Fugu and Fugu Ultra, nothing else. Two exclusions catch people out. Sakana Namazu is pay-as-you-go only — if your work includes the Japanese-specialized model, budget its per-token rates separately. And Fugu Cyber is doubly outside plan territory: it’s access-gated behind a request form, and it requires pay-as-you-go billing mode outright.

The standard Sakana account rules apply to every tier: supported regions exclude the EU/EEA, UK, and Switzerland, and users must be 18 or older. And note what a plan doesn’t change — Sakana’s terms, including the default use of API inputs and outputs for training (there’s an opt-out), apply the same whether tokens come from a plan or from credits.

Before subscribing, it’s worth thirty seconds of orientation: know that orchestration usage exists (it’s why per-token budgeting is hard, and thus why a flat fee is attractive), skim the full pricing picture so you know what the pay-as-you-go alternative costs, and if you’ve never touched the models at all, spend a few cents in the playground first — it’s cheaper than a month of anything.

When to move up (or down) a tier

Because the allowance is defined only relatively, tier decisions are best made from your own experience rather than arithmetic. The signals are straightforward.

Upgrade from Standard when you’re rationing — choosing not to ask Fugu something because you’re worried about the allowance. Rationing a $20 tool with your own attention is a bad trade; Pro’s per-unit price is half of Standard’s, so heavy users are structurally better off there.

Choose between Pro and Max on volume alone. They carry the same per-unit rate; Max simply doubles the ceiling for double the price. If you’re bumping into Pro’s allowance in most months, Max is the same deal with headroom. If you’re not sure, start at Pro — moving up later costs you nothing you’d have kept.

Step down when a month ends with obvious slack. There’s no credit inventory or expiry to manage on plans, so right-sizing is a one-click decision each month rather than a sunk-cost negotiation.

One honest unknown to plan around: Sakana doesn’t publish what happens when you exhaust an allowance mid-month — whether requests throttle, queue, or stop. Assume you may need a fallback (pay-as-you-go credits cover it) if your work can’t pause.

What Sakana hasn’t published about plans

An honest guide should list its unknowns, because they’re exactly where budgeting assumptions go wrong. As of August 2026, Sakana’s public pages do not answer: what the Standard baseline equals in tokens, requests, or dollars; what happens when an allowance runs out mid-month (throttling, queueing, or a hard stop); whether unused allowance rolls over (the relative definitions suggest not, but it isn’t stated); and how plan usage interacts with reasoning-effort settings — higher effort clearly does more internal work, but how that draws down an allowance is undefined publicly.

None of these unknowns is unusual for a subscription this new, and none changes the basic advice above. But if any of them is load-bearing for your decision — say, your work can’t tolerate a mid-month stop — resolve it against your own account’s terms before relying on it, and treat any third-party page (including this one) that claims a specific answer with suspicion until Sakana publishes one.

Plans in practice: the coding-agent pairing

Sakana’s subscription tiers make the most sense in the context they were clearly designed for: the official coding-agent integrations. A one-line installer wires Fugu into Codex or Claude Code, and from then on your terminal sessions run against your plan’s allowance — a flat monthly fee for what is effectively an AI pair programmer that escalates to multi-agent reasoning when a problem is hard.

That pairing is also the best way to discover your real usage tier. A week of normal work inside a coding agent tells you more about whether Standard’s baseline fits you than any amount of forecasting — and if it doesn’t, the upgrade math above takes thirty seconds.

The one workload plans are consistently wrong for: anything automated. Scripts, pipelines, CI jobs, and batch processing belong on pay-as-you-go — they’re latency-sensitive in aggregate, they scale past any allowance, and the documented priority rule works against plan traffic. Estimate that side of the house with the cost calculator instead.

The cheapest ways to try Fugu, ranked

1. Sakana Chat: since August 13, 2026, Sakana’s consumer chat offers Fugu to logged-in users — the most frictionless first taste, no API setup involved. Sakana hasn’t published the terms or limits of that access, so check what you’re offered at login. Details in our chat guide.
2. A few dollars of third-party credit: Fugu Ultra is on OpenRouter at Sakana’s published rates — our browser playground uses exactly this with your own key, showing cost per message. Good for testing API behavior without a Sakana account.
3. Standard plan, one month: $20 for a real working allowance across both models. Cancel if it doesn’t stick.

And remember the availability rules that apply everywhere: no EU/EEA, UK, or Switzerland, and accounts are 18+.

Full rate tables, credits fine print, and worked examples: the complete Fugu pricing guide. New to the model itself? Start at What is Sakana Fugu?