One model.
Three tiers.
No model picker, no API key, no provider account. Eshlon ships one model at three levels of effort, and choosing between them is the only decision you get.
The model selector in the composer showing the three tiers.
16/9Why one model rather than twenty
A model picker is a question most people cannot answer
Offering twenty models looks like generosity and functions as an exam. Which is better at long documents? Which handles a six-hour swarm without drifting? Nobody outside the field knows, and getting it wrong shows up as the product being bad rather than the choice being wrong. Eshlon answers the question once.
Three tiers, all of them strong
The tiers differ by how much reasoning effort the model spends, not by dropping to a weaker model underneath. The cheapest tier available was deliberately skipped: a bottom rung that produces disappointing work makes the whole ladder untrustworthy. Every tier here is one we would ship on its own.
What we will not claim
We did not train it. Not disclosing which model sits underneath is normal and we do the normal thing — but claiming to have trained something is a checkable statement, and a false one is discovered on a schedule rather than by luck. When there is a model we actually trained, that page will say so.
FAQ
Can I bring my own API key?
No. One model, included. There is no provider account to open and no key to manage.
What is the difference between the tiers?
How much reasoning effort is spent per step. Higher tiers think longer on hard problems and cost more.
Did you train Eshlon 1.0?
No, and we are not going to imply otherwise. It is a name for the model Eshlon ships with.
Does it work offline?
The application runs on your machine and your files stay there, but the model is served over the network. No connection, no model.