Pollo Utility BETA

Pollo Assistance's own model: three modes, a real Linux console and agents.

How it differs from the rest

The three tiers — Palos, Luces and Summum — pick the best AI from a pool of third-party models for each message. Pollo Utility is ours: it runs on our own infrastructure, with our personality and our rules, and it can do things a normal chat cannot — open a Linux console, write files you download, and hand work to agents.

The three modes

You pick them from the same dropdown in the chat, next to the model. They are not one engine with longer or shorter answers: each mode is a different model, chosen for what it is being asked to do.

⚡

Fast

Short, to-the-point answers, two to five sentences. It searches the web when it is not sure of a fact. This is the mode we trained ourselves, and the cheapest in credits.

🧠

Reasoning

Developed answers, with context and nuance. It searches the web and also opens the pages when the search snippets are not enough. This is the balance: clearly better than Fast without the cost of Complex.

🔬

Complex

It thinks before answering and shows you that reasoning point by point. It is the only one that can launch agents. For building things and multi-step problems, not for a one-off question.

 FastReasoningComplex
Web search✓✓✓
Open a full page✗✓✓
Linux console✓✓✓
Shows its reasoning✗✗✓
Agents✗✗✓
Credits per token61016
About the console and the three modes. The tool is available in all three, but the one that really uses it — and the one it was designed for — is Complex: opening it and working in it takes several rounds, and that needs a mode built for several rounds. Fast will open it once in a blue moon, for a calculation.

The Linux console NEW

When you ask it to build something, Pollo Utility does not paste a block of code for you to copy: it opens a console, writes the file and hands it to you.

It runs inside your own browser. It is a real Linux, emulated in your tab. It is not a server of ours, it has no internet access, and what is inside belongs to that chat and to nobody else. Come back tomorrow and your files are still there.

You do not see the console, you see the result: files appear under the answer with a download button, and what it did inside shows up as thinking steps — "Wrote index.html", "Calculated", "Delivered 1 file". The console window is hidden on purpose: it is a tool for the model, not another screen for you to watch.

Still in testing, and it shows. A calculation or a single file come out fine. A whole multi-file site works some times and only halfway others. If what you get is not what you expected, say so in the same chat: it reads back what it wrote and fixes it.

Agents NEW

For multi-step work, Complex can delegate. An agent receives a self-contained assignment, works on its own — it has the console and web search — and returns a report. You will see it in the thinking panel as "Used 1 agent".

The agent does not see the conversation, only its assignment. That is what makes it cheap and also its limit: if the assignment is written badly, the agent works blind. That is why only Complex can launch them — writing a good assignment is the hard part.

Thinking, point by point

Complex is the only one of the three that reasons before answering, so it is the only one where you will see this: while it thinks, the header tells you which point it is on and how long it has been going. When it finishes it collapses — "Pollo Assistance thought for 45 seconds" — and you can open it to read the whole reasoning with the steps interleaved in the real order: what it searched, what it ran, what it delivered.

Those steps are not the model telling you what it did: the app adds them when something actually happens. If it claimed to have searched for something and had not, the step would not be there.

What it costs

Pollo Utility spends credits like everything else, but the modes do not cost the same: the table above is credits per token. Complex spends considerably more than the other two, because every round of reasoning is paid for. If you are short on quota, Fast and Reasoning last far longer on the same credits.

🐔 Try Pollo Utility See credits and plans