Fugu Ultra is the higher-performance model in Sakana AI's Fugu family, released on June 15, 2026. Unlike a standard monolithic language model, it is a learned multi-agent orchestration system trained to dynamically route tasks across a swappable pool of underlying models and recursively coordinate instances of itself.
Fugu Ultra is grounded in two ICLR 2026 papers — TRINITY and Conductor — which show how a system can learn to assemble, route, and coordinate expert agents for each task without relying on hand-designed workflows.
Fugu Ultra coordinates expert agents to catch bugs other models miss. Users report it surfaces 20+ issues where other tools find only 3, making it the go-to model for thorough code review.
From a single instruction, Fugu Ultra autonomously reads a paper, implements it, runs training, evaluates results, and analyzes gaps — for hours, without supervision.
Given a scoped instruction, Fugu Ultra drives a full security assessment end-to-end: recon, XSS/SQLi checks, auth review, and a structured report with evidence and retest steps — staying in scope throughout.
Fugu Ultra runs hundreds of experiments autonomously, adapting hyperparameters and training recipes to find the best result, outperforming frontier models on ML research benchmarks.
Mapping 20+ papers and patents — normally 3–4 days of work — can be completed in hours, including connections between papers a human researcher might miss.
Fugu Ultra vs. the Alternatives
- Fugu Ultra: Maximum answer quality for hard, high-stakes tasks. Coordinates the deepest pool of expert agents. Best when accuracy and depth matter more than speed.
- Fugu (standard): Balances strong performance with low latency. Ideal for everyday coding, interactive chat, and code review where response speed matters.
- Single frontier models: Fugu Ultra matches or exceeds GPT-5.5, Opus 4.8, and Gemini 3.1 Pro across coding, reasoning, and scientific benchmarks — without single-vendor dependency.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Fugu Ultra is documented as suited for Kaggle competitions, where its deeper expert-agent orchestration is used to tackle hard, high-stakes optimization problems.
It is also documented for paper reproduction, where users rely on it to reconstruct and verify complex research workflows and results.
Another stated use case is cybersecurity analysis, with early users applying it to demanding security work and complete cybersecurity assessments.
It is further documented for literature and patent investigations, where it is used for deep document-oriented research and analysis.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "sakana/fugu-ultra", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "sakana/fugu-ultra", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"sakana/fugu-ultra","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Fugu Ultra This page | Reasoning + agents | |||
| Reasoning + agents | ||||
| Balanced coding + agents | ||||
| Long-context, multimodal & agentic workflows | ||||
| Reasoning + agents |
Fugu Ultra has a 1,000,000 tokens context window and can return up to 128,000 tokens.
Fugu Ultra takes image, text as input and returns text.
Use sakana/fugu-ultra as the model id. Requests go to https://api.aimlapi.com/v1/chat/completions.
Fugu Ultra became available on June 24, 2026.
Fugu Ultra is priced at input $6.877 / 1M tokens, output $41.262 / 1M tokens, cached input $0.6877 / 1M tokens.
Yes, Fugu Ultra can stream responses as they are generated.
Yes, Fugu Ultra accepts image input alongside text.
Fugu Ultra was built by Sakana AI.
