Chat and pricing guide

Sakana Fugu chat: what is free, what is paid, and what a chat costs

Short answer: Sakana Chat, Sakana AI's consumer chat product, is free with its own usage limits. Chatting with Sakana Fugu models from your own app goes through the API, which has no permanent free tier: subscriptions start at $20 a month, or you pay published token rates on pay-as-you-go. This page separates the two products, shows how an API chat is billed, and lets you price a real conversation below.

Published October 7, 2026 Facts reflect the September 28, 2026 source check

1. "Sakana Fugu chat" can mean two different products

People who search for Sakana Fugu chat usually want one of two things. Some want to open a chat window and talk to Sakana AI's assistant without paying. Others want to put a Fugu model behind their own chat interface, support bot, or coding assistant. Sakana AI sells these as separate products with separate rules, and most confusion about price comes from mixing them up.

The first is Sakana Chat, the consumer chat product. Official pages describe it as free, with its own usage allowance, and with no subscription plan to buy. The second is the Sakana AI API Platform, where the Fugu family lives: fugu, fugu-ultra, fugu-max, and the access-gated fugu-cyber. The API is paid. This site is an independent guide; it does not verify which models Sakana Chat uses internally, so nothing below assumes that Chat and the API share limits, models, or quality.

So the honest answer to "is Sakana Fugu free?" is: the consumer chat product is free, the API is not. Paying for one never changes the other.

2. Sakana Chat versus the Sakana Fugu API

Question	Sakana Chat	Sakana Fugu API
Price	Free, with its own usage limits	Standard $20, Pro $100, Max $200 per month, or pay-as-you-go token rates
Subscription plan	None offered	Three monthly tiers; Pro is 10x and Max is 20x Standard usage
Models you can name	Not documented on the pages this site checks	Fugu, Fugu Ultra and Fugu Max on every tier; Fugu Cyber only on pay-as-you-go after an approved access request
Effect of buying an API plan	No change to Chat limits	Budget cap for API usage; the exact Standard token allowance is not published
Priority	Not documented	Pay-as-you-go tokens are served at higher priority than monthly-plan tokens
Region	The official product page restricts availability in the EU and EEA; see EU availability

Watch one naming collision. The $200 Max plan is a subscription tier, while Fugu Max is a model that every tier includes. A $200 invoice does not mean Fugu Max usage. The full plan comparison is in subscription versus pay-as-you-go.

3. Building your own chat on the Sakana Fugu API

If you want a chat experience inside your product, you call the API. Sakana AI recommends the Responses API for new integrations. The request looks like an OpenAI request, but a few fields behave differently, and two of them matter directly for chat:

No server-side conversation state. previous_response_id is not accepted, so every turn must resend the conversation history in input. Long chats therefore grow the input token count on every turn.
temperature is accepted but ignored on Responses. Do not build a "creativity" slider on it.
max_output_tokens on Ultra limits the final answer only, not the orchestrator's own work, so it is not a complete spend cap.
Pick the model per job. The default fugu is positioned for interactive chat and coding at low latency, fugu-ultra for hard multi-step reasoning, and fugu-max for cost-sensitive work at flat rates.
Pin a version when answers must be reproducible. The moving fugu-ultra alias defaults to v2.0 today and can move; see pin model IDs.

Client setup, base URL handling and streaming are covered in the API quickstart. If a chat call fails, the troubleshooting tree separates key, model ID, region and cost problems before you retry.

4. What one API chat session actually costs

Because the API does not keep conversation state, the cost of a chat is not "messages times a price". Each turn pays for the whole history again as input, plus the new answer as output. Cached input is cheaper when the provider reports it, and the full input total already includes the cached part.

The published standard rates on September 28, 2026 were: Fugu Ultra $5 input, $0.50 cached input and $30 output per million tokens up to 272K request context, rising to $10 / $1.00 / $45 above 272K. Fugu Max is flat at $2 input, $0.25 cached and $6 output at any context length, plus $0.007 per web_search or web_fetch call. The default fugu model is billed by route, so there is no single rate to plug in.

Ultra adds one more line that surprises people: orchestration tokens. The usage object reports orchestration_input_tokens, orchestration_input_cached_tokens and orchestration_output_tokens outside the ordinary totals, and official pricing bills them at the same category rates. A chat app that only logs input_tokens and output_tokens will under-report Ultra spend. The orchestration token guide shows the full formula.

A worked example: a ten-turn support chat that averages 60,000 tokens of resent history per turn, 15,000 of them cached, with 800 output tokens per answer, sends 600,000 input tokens (150,000 cached) and 8,000 output tokens across the session. On Fugu Max that is $0.90 of uncached input, about $0.04 of cached input and about $0.05 of output, so roughly $0.99 before tool calls. On Ultra the same visible usage costs more, and orchestration tokens are added on top. Use the calculator for your own numbers instead of trusting a rule of thumb.

5. Price a real conversation

Paste the totals from your API responses for one chat session. Enter the full input total including cached tokens, then the cached part once. This is the same published-rate calculator as the calculator page; it covers Fugu Ultra and Fugu Max only, because Cyber rates are not public and Fugu's rate depends on the route.

Model Rate tier for this scenario

Fugu Ultra charges one rate up to 272K request context and a higher rate above it.

User-visible usage

Enter the full input_tokens total from the API response: it already counts cached_tokens. List the cached part once as well, because the calculator bills the rest at the input rate and the cached part at the cached rate. Do not add cached tokens on top of the total.

Input tokens (including cached) Cached input tokens Output tokens

Orchestration usage

These fields are extra Ultra usage outside the ordinary request totals above, not a breakdown of them. orchestration_input_tokens likewise includes orchestration_input_cached_tokens. Leave them at zero to cost only the visible request and response.

Orchestration input (including orchestration cached) Orchestration cached input Orchestration output

Tool calls

Fugu Max publishes $0.007 per web_search or web_fetch call. No per-call rate is published for Ultra, so this field must stay at zero for Ultra.

web_search / web_fetch calls

6. Which one should you use?

You just want to chat for free

Use Sakana Chat from Sakana AI's official site. It is free with its own limits, and no API plan changes those limits.

You are prototyping a chat feature

A Standard or Pro API subscription works as a budget cap for light, interruptible use. You still cannot forecast remaining tokens from the public pages.

Your chat is in production

Official copy points heavy production work at pay-as-you-go because those tokens are served at higher priority. Estimate with the calculator above, including orchestration on Ultra.

Your chat does security research

Fugu Cyber is available only on pay-as-you-go after an approved access request, and its token rates are not published. See Cyber access.

Your users are in the EU or EEA

Stop. The official product page restricts availability there; do not proxy around it.

Frequently asked questions

Is Sakana Fugu chat free?

Sakana Chat, Sakana AI's separate consumer chat product, is free with its own usage limits. The Sakana Fugu API has no permanent free tier: paid subscriptions start at $20 a month, or you pay published token rates on pay-as-you-go.

Does a Sakana Fugu API subscription raise my Sakana Chat limits?

No. The $20, $100 and $200 tiers are API Platform subscriptions. Sakana Chat has no subscription plan, and buying an API subscription does not raise or remove its limits.

Can I chat with Fugu Ultra or Fugu Max through the API?

Yes. Every API tier includes Fugu, Fugu Ultra and Fugu Max, and a multi-turn chat is built by sending the full conversation history in each request, because the Responses API does not accept previous_response_id.

How much does one API chat session cost?

Count the tokens from the usage object and apply the published rates: Fugu Ultra is $5 input, $0.50 cached and $30 output per million tokens up to 272K context, and Fugu Max is a flat $2 / $0.25 / $6. Ultra also bills orchestration tokens on top. The calculator on this page does the arithmetic.

Is Sakana Chat or the Sakana Fugu API available in the EU?

The official product page restricts availability in the EU and EEA. Treat that as a stop for production users there rather than routing around it.

Sources and verification boundary

Sakana AI Pricing documentation: plan prices, pay-as-you-go priority, Ultra and Max rates, orchestration billing
Sakana AI Models documentation: model IDs, Responses API field behavior, Cyber access rule
Sakana AI Fugu product page: product scope and regional availability

Prices and rules here reflect the September 28, 2026 recheck. This site has not independently tested Sakana Chat's limits or models. Recheck the official pricing page before a 2026 budget decision.