Fugu Ultra v2 pricing in 2026

What one task costs on Fugu Ultra v2

Per-million-token prices are hard to feel, so here are typical token counts for common jobs priced with Fugu Ultra v2. Figures are USD at the rates above and exclude prompt caching and batch discounts.

TaskTokens (in / out)Cost
Code review of a 500-line pull request60,000 / 4,000$0.420
Debug one failing test40,000 / 6,000$0.380
Generate a landing page in one shot3,000 / 30,000$0.915
Draft a 1,200-word article2,000 / 2,500$0.085
Answer a question from a 20-page document25,000 / 800$0.149
Multi-step agent run (10 tool calls)250,000 / 25,000$2.00
All of the above, once$3.95

Cheaper at the same or better capability: GLM-5.3 FlashX runs the same set of tasks for $0.226 instead of $3.95.

Fugu Ultra v2 by Sakana AI costs $5 per 1M input tokens and $30 per 1M output tokens — about 42.1k tokens per $1. Prices last verified 2026-09-19.

The verdict

The higher-performance tier of Sakana AI's learned multi-agent orchestration system, at $5 / $30 per 1M tokens. Like Fugu Max it routes work across a team of agents rather than answering with one model. Artificial Analysis has not benchmarked it, so it carries no Intelligence Index here — the rating shown is a tier estimate, not a measurement.

A concrete example: sending 100,000 input tokens and getting back 20,000 output tokens with Fugu Ultra v2 costs about $1.10 at current prices.

Who is Fugu Ultra v2 for?

Fugu Ultra v2 is the higher-performance member of Sakana AI's Fugu family — a learned multi-agent orchestration system that assembles a team of agents to work a problem, rather than scaling a single model. At $5 per 1M input and $30 per 1M output it is priced in premium territory, and above 272k prompt tokens the rate doubles.

As with Fugu Max, the limitation on this site is data: there is no Artificial Analysis Intelligence Index for it, so it cannot be ranked on intelligence per dollar, and the figure shown is a tier-based estimate. If you are evaluating it, do so against its own published benchmarks — we have kept the entry so the catalogue is complete, not because we can vouch for its ranking.

Strengths & weaknesses

  • Strength: Highest-capacity Fugu tier for hard multi-step tasks
  • Best for: Agentic, Quality
  • Weakness: Unbenchmarked on Artificial Analysis; premium pricing without a verified index.

Fugu Ultra v2 key facts

ProviderSakana pricing
Input price$5 / 1M tokens
Output price$30 / 1M tokens
Tokens / $142.1k
Intelligence / $133,684
Context window1.0M tokens
Tierbalanced
SpeedSlow
Released2026-09-11
Last verified2026-09-19

What Fugu Ultra v2 delivers for your budget

For a coding workload, here is roughly what each monthly budget buys:

BudgetTokens / monthMonths of workload
$201.9M2.9x
$504.6M7.1x
$1009.3M14.3x

Other Sakana models

How Fugu Ultra v2 compares to its closest rivals

Rather than repeating the full catalog on every page, here are the three current models nearest to Fugu Ultra v2 on the intelligence index — and why you might still pick one instead:

  • GLM-5.3 FlashX (GLM) — $0.37 input / $1.25 output per 1M tokens. Why you might pick it: GLM-5.3 quality at high throughput, with a first-party price.
  • Fugu Max (Sakana) — $2 input / $6 output per 1M tokens. Why you might pick it: Multi-agent orchestration as a single endpoint.
  • GPT-6 Astra (ChatGPT) — $10 input / $50 output per 1M tokens. Why you might pick it: Frontier reasoning inside ChatGPT Plus and Pro.

Fugu Ultra v2 — frequently asked questions

How much does Fugu Ultra v2 cost?

Fugu Ultra v2 by Sakana AI costs $5 per 1M input tokens and $30 per 1M output tokens.

How many tokens per dollar does Fugu Ultra v2 give you?

On a typical 1:3 input-to-output mix, Fugu Ultra v2 delivers around 42.1k tokens per dollar — an intelligence-per-dollar score of 33684.

What is Fugu Ultra v2's context window?

Fugu Ultra v2 supports a context window of 1.0M tokens.

Is Fugu Ultra v2 worth it in 2026?

Fugu Ultra v2 is Sakana's balanced tier model. Use the AI Intelligence Calculator to see how its value compares to every other model at your budget.

Advertisement