You choose the tier, in the composer, before you send. Inside that tier Esan spreads the work across several models by role — the strong one plans and writes, the light one handles the routine steps — so a tier is never a single model.
Fast and light
The quick one. Instant answers, short drafts, simple lookups and the routine steps inside a bigger task. Low cost, so it does the everyday work without you thinking about it.
Best for
The everyday workhorse
Strong reasoning for real work — research, coding, analysis, planning. The default for most tasks: capable enough to be trusted, efficient enough to run all day.
Best for
The most capable
Reserved for the hardest, longest, highest-stakes work — deep reasoning, long autonomous runs, and the moments where getting it exactly right matters more than speed.
Best for
One task is dozens of model calls, and they are not all worth the same money. Esan routes them by role rather than sending everything to the model you picked.
The model that plans the task and writes the final answer is the one the tier is named for. That is the part you feel: how well the work is thought through, and how good the writing is at the end.
Routing, summarising, naming a chat, deciding which tool to call next — the dozens of small steps inside one task run on the light model no matter which tier you picked. You are not billed premium rates for bookkeeping.
The sub-agents that read code, search the web and reason at length are held to the two stronger models. A specialist that cannot do its job is worse than no specialist, so that floor does not move.
Next to the send button, on every chat. Change it before you send and the task runs on the new tier — there is no settings page to visit and nothing to save.
A chat keeps the tier it was last sent with, so a long piece of work stays where you put it while a quick question elsewhere stays cheap.
Which tiers you can reach depends on your plan. See plans →