Sakana Fugu Ultra: Japan's New $5/M Input Model — Coding Cost Analysis
June 29, 2026 · 7 min read
What Is Fugu Ultra?
On June 22, 2026, Sakana AI — a Tokyo-based research lab founded by former Google Brain researchers — released Fugu Ultra (model ID fugu-ultra-20260615). It is available via the Sakana API, the Vercel AI Gateway, and OpenRouter under sakana/fugu-ultra.
The published pricing from the official Sakana product page at sakana.ai/fugu:
| Tier | Input ($/M) | Output ($/M) | Cached Input ($/M) |
|---|---|---|---|
| Standard (≤272K context) | $5.00 | $30.00 | $0.50 |
| Long context (>272K) | $10.00 | $45.00 | $1.00 |
The 1M-token context window is real, but using it above 272K doubles the input price — a tiered structure worth knowing before you load a large codebase into context.
How It Sits in the Pricing Tier
Fugu Ultra's standard tier is essentially price-parity with Claude Opus 4.8 on input, but substantially more expensive on output:
| Model | Provider | Input ($/M) | Output ($/M) |
|---|---|---|---|
| Fugu Ultra | Sakana | $5.00 | $30.00 |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 |
| GPT-5.5 | OpenAI | $5.00 | $30.00 |
| GPT-5.4 | OpenAI | $2.50 | $15.00 |
| Grok 4 | xAI | $3.00 | $15.00 |
Fugu Ultra and GPT-5.5 share an identical rate card. Claude Opus 4.8 is $5 cheaper on output per million tokens. GPT-5.4 and Grok 4 cost meaningfully less and cover many of the same use cases.
Coding Cost Estimate: Medium Project
Using our standard estimator parameters — a medium project (15,000 LOC), CLI agent tooling, production quality — the token totals are approximately 120M input tokens and 18M output tokens for the full build.
| Model | Input Cost | Output Cost | Total |
|---|---|---|---|
| Fugu Ultra | $600 | $540 | $1,140 |
| Claude Opus 4.8 | $600 | $450 | $1,050 |
| GPT-5.5 | $600 | $540 | $1,140 |
| GPT-5.4 | $300 | $270 | $570 |
At standard pricing, Fugu Ultra and GPT-5.5 are identical in total cost. Claude Opus 4.8 is $90 cheaper on the same project purely from the $5/M output advantage. GPT-5.4 is half the price if you can accept a tier-down in reasoning capability.
The Long-Context Trap
Fugu Ultra's 1M context window is genuinely useful for large codebase ingestion — but the tiered pricing changes the calculus when you're above 272K tokens. At the long-context rate ($10 input, $45 output), the same medium project estimate doubles in cost to roughly $2,000.
CLI agents like Claude Code routinely push 150K–400K input tokens per turn on large codebases. Developers loading a 50,000+ LOC repo into Fugu Ultra should assume they will hit the long-context tier regularly, which changes the break-even analysis substantially.
When to Consider Fugu Ultra
Fugu Ultra is worth evaluating in two scenarios. First, if you need a non-US provider for compliance, data residency, or supply-chain diversity — Sakana is a Japan-headquartered lab, which may matter for enterprise procurement decisions. Second, if you're already using OpenRouter and want a turnkey alternative to Claude Opus 4.8 or GPT-5.5 with the same price card and compatible API format.
For pure cost optimization within this tier, Claude Opus 4.8 at $25 output wins over Fugu Ultra at $30. If your project stays under 272K context per turn, the gap is $90 on a medium project — small but real. If you're regularly going long-context, the gap blows out and Fugu Ultra becomes the most expensive option in this tier.
Want to calculate exact costs for your project?
Frequently Asked Questions
What is Sakana Fugu Ultra?
Fugu Ultra is a large language model from Sakana AI, a Tokyo-based research lab. It's available via the Sakana API, Vercel AI Gateway, and OpenRouter. Standard pricing is $5 input / $30 output per million tokens with a 1M context window.
How does Fugu Ultra compare to Claude Opus 4.8 in price?
Both have $5/M input. Fugu Ultra charges $30/M output vs Claude Opus 4.8's $25/M. On a medium coding project, that works out to roughly $90 more expensive for Fugu Ultra at standard context lengths.
Does Fugu Ultra's 1M context window cost extra?
Yes. For prompts exceeding 272K tokens, Fugu Ultra switches to a long-context tier: $10 input and $45 output per million tokens — double the standard rate.
Is Fugu Ultra worth using for AI coding?
For short-to-medium context coding tasks, it's cost-equivalent to GPT-5.5 but slightly more expensive than Claude Opus 4.8. Its main advantage is provider diversity (Japanese lab, non-US infrastructure), which matters for some enterprise compliance requirements.
Related Articles
Fugu Ultra vs Claude Opus 4.8 vs GPT-5.4: Which $5/M Model Is Best for Coding?
Three models cluster near the $5/M input price point: Sakana Fugu Ultra ($5/$30), Claude Opus 4.8 ($5/$25), and GPT-5.4 ($2.50/$15). We compare them on coding cost efficiency, context pricing, and when to use each.
OpenRouter Launches MCP Server: One-Click Model Comparison Without Leaving Your Coding Agent
OpenRouter released an MCP server giving coding agents real-time access to model pricing, benchmark scores, and documentation. We walk through what it does, how to install it in Claude Code or Cursor, and how it changes day-to-day model selection workflow.
Claude Fable 5 Pricing: $10/$50 Per Million Tokens — Is Anthropic's Strongest Model Worth It for Coding?
Claude Fable 5 launched at $10 input / $50 output per million tokens — less than half of Mythos Preview pricing. We analyze when the premium over Opus 4.8 at $5/$25 is justified for coding workflows.