---
title: Models & routing
description: IsonAI's model tiers — Adaptive, Omni, LYT, and PRME — how automatic selection works, explicit selection, and fallback behavior.
last_verified: 2026-07-17
---

# Models & routing

IsonAI offers several **model tiers** with different capabilities and speeds. You can let IsonAI choose automatically (Adaptive) or lock a specific tier from the model picker in the chat top bar.

Terms used in these docs: **tier/model** is the engine that answers (Omni, LYT, PRME); **mode** is a behavior setting on top of it (Adaptive for tier selection, [Thinking](/en/thinking/) for reasoning depth).

## Adaptive (default)

Adaptive automatically picks the most suitable tier for each question — fast for simple things, deeper for complex ones. Most users can just use Adaptive and never think about model choice.

Weighty or high-stakes topics — legal, financial, medical, tax, scam detection, in-depth comparisons, and coding — are automatically routed to a stronger tier.

## IsonAI Omni

The main tier for complex tasks: layered analysis, long reasoning, thorough comparisons, structured writing, and [agentic behavior](/en/tools/) such as search-read-conclude across sources. Omni is also the engine behind [Naya](/en/naya/) and [Deep Research](/en/deep-research/).

## IsonAI LYT

The lightweight tier that prioritizes speed — for casual conversation, greetings, and simple everyday questions. LYT **does not use the daily quota**, so it stays available when the smart-model quota runs out. Its capabilities are more limited: web search is more concise and there is no deep reasoning. On LYT, [Thinking](/en/thinking/) always behaves as **Fast**.

## PRME

The coding-specialist tier — strong at writing and analyzing code, mathematics, and structured technical analysis. PRME is **only active when selected explicitly**; Adaptive never routes questions to PRME on its own. PRME is also the engine behind [IsonForge](/en/isonforge/).

## Explicit selection

Your choice in the model picker is **always respected**. If you lock Omni, every question is answered by Omni; the same goes for LYT and PRME. The only exceptions occur when quota runs out or a tier is temporarily unavailable (see below), and the change is always visible — never a silent substitution.

## Fallback and degradation

- **Quota exhausted** — when the daily smart-model quota runs out, answers automatically continue on LYT until the quota renews, with a notice in the interface. See [Plans & quota](/en/plans/).
- **Tier outage** — if the main tier is temporarily unavailable, IsonAI switches to another healthy tier when possible to keep the service running.
- **PRME** — as an explicit tier, PRME shows a clear error message when its engine is unavailable, rather than silently answering with another tier.

## Choosing the right tier

| Need | Choice |
|---|---|
| Everyday use | Adaptive |
| Deep analysis, long documents, important decisions | Omni |
| Light chat as fast as possible, quota-friendly | LYT |
| Writing or dissecting code | PRME (or Adaptive, which routes coding to Omni) |

## Related

- [Thinking](/en/thinking/) — reasoning depth on top of any tier.
- [Plans & quota](/en/plans/) — daily quota per tier.