by SakanaReleased 11 September 2026
Fugu Ultra v2 reads text, pictures and PDFs and writes text, keeping about 1,000 pages in mind at once. It is a team of models run by one that hands each task to the right helper, tuned for answer quality on hard multi-step reasoning, independent research and full software projects, with built-in web search and the exact tables you ask for.
You pay for what each request uses. No subscription, no minimum.
One message, the ordinary use
Every rate
| Meter | ↑ you send | ↓ comes back |
|---|---|---|
| Text tokens | 1 664,18 DA | 9 985,07 DA |
| Cached tokens read | 166,42 DA |
per 1M tokens
Web search · 3,33 DA per call, under every choice
One company runs this model. OpenDunes sends every request to it. DA per 1M tokens: ↑ what you send · ↓ what comes back.
Also charged: prompts over 272 000 tokens ↑ 3 328,36 ↓ 14 977,60 · a web search 3,33 DA each.
You pay the price of the provider that served your request. Prices read 15:21 · speed over the last 30 minutes · uptime over 3 days.
Real requests, real replies. Try any of them yourself — one click, pre-filled.
What's the total on this invoice, and when is it due?
invoice-scan.jpg · photoThe total is 42 300,00 DA, due 5 September 2026. Worth a double-check: the handwritten day could read as 8 instead of 3.
Try in Chat LabTurn this customer email into JSON with intent, urgency and order_id.
customer-email.txtTry your own request — same model, pre-selected.
Open Chat LabSwap the base URL and the key. Your OpenAI code runs unchanged.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://opendunes.com/v1",
api_key=os.environ["OPENDUNES_API_KEY"],
)
resp = client.chat.completions.create(
model="sakana/fugu-ultra-v2",
messages=[{"role": "user", "content": "Two lines to welcome a shop in Oran."}],
)
print(resp.choices[0].message.content)Runs this request once on your balance.
Standard OpenAI fields, grouped by what they control. Anything not listed is accepted and ignored.
Reasoning
Let it think before answering — and see the thinking.
reasoningreasoning_effortinclude_reasoningStructured output
Get JSON you can parse instead of prose.
structured_outputsTools
Let the model call your functions.
toolstool_choiceOther
Also accepted by this model.
web_search_optionsAbout 6,66 DA per message. ↑ 1 664,18 DA · ↓ 9 985,07 DA per 1M tokens.
No. You pay per use from your balance, with no subscription. Try it in Chat at the same price.
Fugu Ultra v2 reads text, pictures and PDFs and writes text, keeping about 1,000 pages in mind at once. It is a team of models run by one that hands each task to the right helper, tuned for answer quality on hard multi-step reasoning, independent research and full software projects, with built-in web search and the exact tables you ask for.
Fugu Ultra costs the same for the same message.
Deposit dinars with your Edahabia or CIB card. Every model, Fugu Ultra v2 included, is paid from that balance. No foreign card, no conversion.
Sakana names no languages for Fugu Ultra v2. Arabic, French and Darja: not stated.
It reads text, images and PDFs. Up to 1M tokens in one request, about 1 000 pages. It writes up to 96 000 words at once.
No. Sakana has not published its weights.
Sakana. Released 11 September 2026. It knows the world up to 28 August 2026.
Fugu Max reads text, pictures and PDFs and writes text, keeping about 1,000 pages in mind at once. It is a team of models run by one that hands each task to the right helper, can search and read the web by itself, and lets you set how hard it thinks. Sakana makes it the balanced choice of its family, trading quality against cost.
Fugu Ultra reads text and pictures and writes text, keeping about 1,000 pages in mind at once. It is a team of models run by one that hands each task to the right helper, aimed at hard reasoning, coding and long jobs that use connected tools, with web search built in and an adjustable thinking effort.
Sakana Namazu is built for Japanese: it starts from Kimi K2.6 and adds training in the Japanese language and in business settings. It reads text, pictures and files, writes text, and thinks before it answers, with about 260 pages in mind at once. It suits following Japanese instructions, business writing, maths, coding, research and long step-by-step jobs. Sakana may use your inputs for training unless you opt out in its console, so choose a general model for work in other languages.
From what its maker and two benchmark sites publish, and where it stands on OpenDunes. Read 7 Nov 2026.
Thinks first
Works through the problem before it answers.
Uses your tools
Can call functions your code provides.
Fills your JSON
Answers in the JSON shape you give it.
Searches the web
Can look things up online while it answers.
Cheaper on repeats
A prompt it saw a moment ago costs less the second time.
Open weights No.
Read on sakana.ai.
Knows events up to 28 Aug 2026
What it learned stops at this date.
Research a question across many sources and write the answer
Thinks first
Build a full-stack feature from a written brief
Thinks first
Read a PDF report and answer questions on it
Files