Sakana Fugu Max API & Pricing Guide

01

Overview

Sakana AI's Fugu Max can be used through the same OpenAI-compatible API surface as the rest of the Fugu family. That matters because teams already using an OpenAI-style client or coding harness can test Sakana's orchestration layer without rebuilding the entire application around a proprietary SDK.

This guide focuses on the practical developer questions: which model IDs to use, what the current token prices are, how Fugu Max differs from Fugu Ultra v2, what extra tool charges exist, and what still needs to be verified before production deployment.

02

Current Fugu model IDs

Sakana's live pricing documentation currently lists:

• fugu-max-v1.0 — the cost-performance-focused Fugu Max model.
• fugu-ultra-v2.0 — the higher-cost Fugu Ultra v2 model aimed at difficult reasoning, research and software-engineering workloads.

Sakana says the API is OpenAI-compatible. Existing Fugu users can move between compatible Fugu models by changing the model parameter rather than redesigning the integration. That is useful for A/B testing because the application layer can stay mostly constant while the orchestration mode changes.

OpenAI compatibility should not be read as behavioral equivalence. Fugu can coordinate multiple expert agents behind the endpoint, so response time, cost and answer structure may differ from a request handled by one foundation model.

03

Fugu Max API pricing

Sakana currently lists Fugu Max at:

• Input: $2 per million tokens.
• Output: $6 per million tokens.
• Cached input: $0.25 per million tokens.

The pricing page says those Fugu Max rates remain fixed regardless of context length. Built-in web_search and web_fetch are separate charges at $0.007 per call.

For a simple example, one million uncached input tokens plus 250,000 output tokens would cost $3.50 at those published rates before any tool calls: $2 for input plus $1.50 for output. That is an arithmetic illustration, not a promise about the total cost of completing a real task. Agentic systems can generate different amounts of output and invoke tools differently depending on the request.

04

Fugu Ultra v2 pricing

Fugu Ultra v2 currently has higher rates:

• Input: $5 per million tokens.
• Output: $30 per million tokens.
• Cached input: $0.50 per million tokens.

For contexts above 272K tokens, Sakana currently lists higher Ultra rates:

• Input: $10 per million tokens.
• Output: $45 per million tokens.
• Cached input: $1 per million tokens.

Using the same one-million-input plus 250,000-output example at standard Ultra pricing produces $12.50 before tools: $5 input plus $7.50 output. Again, that is only a rate calculation; actual request cost depends on the real token mix, caching and tools.

05

Why the pricing difference exists

Fugu Max and Fugu Ultra v2 are not simply two quality settings on one LLM. Sakana describes them as different orchestration products with different model pools and objectives.

Fugu Max is designed to expand the cost-performance frontier by routing work across open-weight and specialist models. Fugu Ultra v2 is aimed at maximizing result quality on harder multi-step tasks. That makes the output-token price difference especially important for long-form research, coding or agent workflows that can generate large responses.

06

OpenAI-compatible integration path

Sakana states that Fugu is exposed through an OpenAI-compatible API and that existing Fugu users can switch models with a one-line model change. In practice, the safe migration process is:

1. Keep your existing application logic unchanged where possible.
2. Select the exact current Sakana model ID.
3. Run representative test prompts against both your current model and Fugu.
4. Compare answer quality, total token use, tool calls, latency and failure rate rather than only comparing list prices.
5. Test long-context behavior separately if you intend to use Ultra above the 272K pricing threshold.
6. Verify provider/privacy requirements before production because Fugu Max and Ultra use fixed model pools.

The exact base URL, authentication details, request fields and any rate limits should be checked against Sakana's current API documentation at implementation time. They are operational details that can change and should not be reconstructed from old examples.

07

Web search and fetch charges

Sakana's current pricing page lists both web_search and web_fetch at $0.007 per call. Those charges sit on top of model token usage.

This means two requests with identical input/output token counts can have different totals if one invokes search or fetch repeatedly. For research-heavy agents, tool calls therefore belong in the cost model rather than being treated as free functionality.

08

Subscription plans vs API token pricing

Sakana's Fugu product page also advertises monthly plans: Standard at $20/month, Pro at $100/month and Max at $200/month. Those product plans should not be confused with the Fugu Max API model itself.

fugu-max-v1.0 is a model identifier with pay-as-you-go token pricing. The $200/month 'Max' plan is a service subscription tier. The shared use of the word Max can be confusing, so production documentation should name the model ID explicitly.

09

Privacy and model-pool caveat

Sakana says the standard Fugu product can let users opt out of specific models in the pool for privacy and compliance reasons. Fugu Max and Fugu Ultra use fixed pools. Teams with provider restrictions or regulated data therefore should not assume they can exclude an underlying provider on their own.

Sakana directs organizations that need custom provider configurations toward enterprise arrangements. Before using Fugu for sensitive workloads, verify the current model-pool, data-handling and contractual controls directly with Sakana.

10

What is not publicly fixed

Do not build production assumptions around numbers Sakana has not published as universal guarantees. The current public material does not establish one universal latency number, request-per-minute quota, minimum monthly API spend or cost per completed task.

Those values can vary with account, prompt, orchestration path and future product changes. The reliable public numbers today are the token and tool-call rates listed in Sakana's live pricing documentation.

11

Fugu Max or Ultra v2 for API workloads?

Fugu Max is the natural first test when you care about throughput economics and want orchestration without paying Ultra's higher output rate. Ultra v2 makes more sense when the value of a stronger result on difficult work outweighs the higher token price.

The best production choice is empirical: run the same representative workload through both, record total cost and success quality, and use the cheaper option only when it meets your quality threshold.

12

FAQ

What is the Fugu Max API model ID?

fugu-max-v1.0 according to Sakana's current pricing page.

What is the Fugu Ultra v2 model ID?

fugu-ultra-v2.0.

Is Fugu compatible with OpenAI clients?

Sakana describes Fugu as OpenAI-compatible. Verify the current endpoint and request schema in Sakana's live docs before deployment.

Does Fugu Max have long-context surcharges?

Sakana's current pricing page says Fugu Max uses fixed rates regardless of context length. Ultra v2 has a higher pricing tier above 272K context.

Are web search calls free?

No. Sakana currently lists web_search and web_fetch at $0.007 per call.

Is the $200/month Max plan the same as Fugu Max API pricing?

No. The monthly Max subscription tier and the fugu-max-v1.0 API model are different concepts.