Fugu Ultra v2: pricing, context and performance
Fugu Ultra v2 in numbers: price per token, context window, independent benchmark scores and comparable models, based on public data checked on October 22, 2026.
Data as of October 22, 2026
Key figures
| Developer | Sakana AI |
|---|---|
| Released | September 11, 2026 |
| Knowledge cutoff | August 28, 2026 |
| Context window | 1,000,000 tokens (≈ 1,500 pages) |
| Maximum output | 128,000 tokens |
| Input price | $5 per million tokens (median of 6 providers) |
| Output price | $30 per million tokens |
| Open weights | No |
| Reasoning mode | Yes |
| Tool calling | Yes |
| Structured output (JSON) | Yes |
| Accepted inputs | text, image, PDF |
| Providers offering it | 6 |
Fugu Ultra v2 at a glance
Fugu Ultra v2 is a language model from Sakana AI, released on September 11, 2026. It is a very recent release. It is a closed model: you access it through an API, from Sakana AI or one of the 6 providers listed on models.dev. It has a reasoning mode: it can "think" before answering, which helps with multi-step problems but uses more tokens, and therefore more money.
We file it under frontier models (24 models in this category on the site).
Benchmark results
As of October 22, 2026, Epoch AI publishes no score for Fugu Ultra v2 on our basket's benchmarks. To place it, rely on its specifications and price, then test it on your own cases.
What it can do
Its context window reaches 1,000,000 tokens, roughly 1,500 pages of text at 500 words per page: enough to load a whole book or a mid-sized codebase. A single answer can run to 128,000 tokens.
It accepts as input: text, image and PDF. Its knowledge stops at August 28, 2026: beyond that, it needs web search or your documents.
It can call tools (functions, APIs) and return structured JSON output, which makes it a possible foundation for an AI agent.
How much does Fugu Ultra v2 cost?
Fugu Ultra v2 costs $5 per million input tokens and $30 per million output tokens (median of 6 providers, checked on October 22, 2026).
To put it in perspective: reading 20,000 tokens (about thirty pages) and writing 2,000 costs roughly $0.16; a thousand of those tasks, $160.
That is 2.1 times the median output price of frontier models ($14.5).
To estimate your own budget, use the agent cost calculator.
Who is it for?
Given its category, Fugu Ultra v2 is mainly aimed at hard tasks where quality matters more than cost: deep reasoning, complex code, long analyses.
Its very large context window makes it worth considering for analysing large sets of documents in a single request.
Before adopting it, test it on your own examples: benchmarks measure standard tasks, not your job.
Alternatives
In the same category, the models closest to Fugu Ultra v2 are:
- GPT-5.5 (OpenAI, index 89)
- Fugu Ultra (Sakana AI, $30 per million output tokens)
- Claude Opus 5 (Anthropic, index 92.7)
- Claude Opus 5.5 (Anthropic, index 95.9)
Similar models
- GPT-5.5 · OpenAI · index 89
- Fugu Ultra · Sakana AI
- Claude Opus 5 · Anthropic · index 92.7
- Claude Opus 5.5 · Anthropic · index 95.9
Other Sakana AI models
- Fugu Max · September 11, 2026
- Sakana Namazu · August 3, 2026
Frequently asked questions
How much does Fugu Ultra v2 cost?
As of October 22, 2026, Fugu Ultra v2 costs $5 per million input tokens and $30 per million output tokens (median of 6 providers). A task that reads 20,000 tokens and writes 2,000 costs about $0.16.
What is the context window of Fugu Ultra v2?
1,000,000 tokens, roughly 1,500 pages of text. An answer can reach 128,000 tokens.
Is Fugu Ultra v2 open source?
No. It is a closed Sakana AI model, available only through an API.
When was Fugu Ultra v2 released?
Fugu Ultra v2 was released on September 11, 2026, according to models.dev.
Can you build an agent with Fugu Ultra v2?
Yes: it can call tools and return structured JSON, the two basic building blocks of an agent.
Data sources
- Models.dev — Model data: models.dev (open-source project)
- OpenRouter — Market prices cross-checked with OpenRouter
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.