Skip to content

Tartarus AI models

Three models, chosen per conversation from the picker. They differ in one thing that matters more than the rest: whether the model reasons before it answers, and what that reasoning is tuned for.


Tartarus AI Fast

Quick answers for everyday questions.

The default. It answers immediately without a thinking pass, which is what you want for drafting, rewriting, summarising, explaining a concept or any question where the first good answer is the right answer. Tuned to stay coherent over long output instead of looping.

Writing and rewritingSummariesExplanationsEveryday questions

Available on: Basic, Standard, Pro, Ultra

Tartarus AI Deep

Thinks before answering. Built for hard problems.

A reasoning model. It works through the problem first and the reasoning is streamed on its own channel into a collapsible panel, so you can read how it got there without that thinking being mixed into the answer. Slower per reply, and worth it when the answer has to be right rather than fast.

Multi-step problemsAnalysis and trade-offsMathsPlanning

Available on: Pro, Ultra

Tartarus AI Balanced

Thinks a little before answering. A good default.

Reasons before it answers, but briefly — between Fast and Deep on both thinking time and patience for a long problem. Sampling is tuned to leave repetition alone, which matters wherever the exact wording is the point rather than something to paraphrase.

Everyday workTechnical questionsStructured writingStep-by-step tasks

Available on: Standard, Pro, Ultra

Side by side

Comparison of the Fast, Balanced and Deep models
FastDeepBalanced
Answers immediatelyYesNo, thinks firstNo, thinks first
Reasoning panelYesYes
Best forEveryday questionsHard problemsSource code
Web searchYesYesYes
Available onAll plansStandard, ProPro

Which model should you use?

Start on Fast. It is the right answer for most of what people actually ask an assistant: draft this, shorten that, explain how something works, what is the difference between these two things. Reasoning costs time and tokens, and on a question with one obvious answer it buys nothing.

Move to Deep when a wrong answer would be expensive and the problem has more than one step: comparing options against constraints, working through maths, planning something with dependencies. Use Balanced for everything in between — it thinks briefly first, and leaves repetition alone, which matters wherever the exact wording is the point rather than something to paraphrase.

See which plans include which models