Replit Auto Mode Now Picks the Model for You
Replit now routes each Agent task to a model automatically, and says tests matched old Max quality at 65% lower cost. Treat Auto as a paid default, not Free Mode, and pin a model before you ship.
Replit has started choosing the model for each Agent task, so you no longer have to.
What happened
On 26/08/2026 Replit switched on Intelligent Model Routing for every account.
When you see Auto in Agent settings, Replit is picking a model for that step of the work, then picking again as the task gets harder or simpler.
The company says its tests delivered the same output quality at 65% lower cost than the previous version of Max Mode.
The 28/08/2026 product changelog made the same routing the default Agent mode for Enterprise workspaces.
On Core and Pro plans, Auto sits inside Power Mode or Max Mode and spends subscription credits like any other paid run.
Free Mode still routes in the background and still will not let you pick a model by hand.
Core and Pro users can open the model selector and pin a specific model when they want that control.
Enterprise admins can limit Auto to an approved set of models per workspace, and they can turn Auto off so people choose from the allowed list themselves.
The Enterprise Agent menu now shows Power and Max only.
Lite has gone, and Power has replaced the old Economy Mode at the same price.
Michele Catasta, president and head of AI at Replit, told The New Stack that per token rates across one model family can span orders of magnitude, while cheaper models have closed much of the quality gap.
That is the commercial case for letting the Agent pick, rather than leaving every task on the dearest model because nobody wants to keep switching.
Why this matters for UK businesses
Most small firms do not have a person whose job is to know whether Claude, Gemini or an OpenAI model is the right choice for a form, a database change or a messy bug.
They want a working internal tool, a predictable bill, and a way to say which systems the company is allowed to use.
Automatic routing helps with the first two.
You describe the outcome, and Replit decides how much model to throw at each step.
The 65% figure will get quoted in budget meetings, and it needs a sober reading.
It is Replit's own test against the old Max Mode, not a promise that your next invoice will fall by two thirds.
The more useful change is that the cost of a task can now move with the work, instead of sitting at Max because switching models by hand felt like a chore.
For teams on Enterprise, the admin control is the part I would take to a finance director.
You can keep Auto on so staff are not shopping for models, and still block a provider that your policy or a client contract will not accept.
How we see it at Adevious AI
I would leave Auto on for paid Power and Max work as the default on the builds I run.
I am a Replit Ambassador, and I have wasted enough mornings picking a model that was going to be the wrong one two prompts later.
Auto is the setting I would actually use for iteration.
The caution is simple: Auto is not Free Mode.
It spends credits.
If you confuse the two, a long Max session on Auto will still chew through the budget, just at a lower rate than the old Max default.
The other limit is visibility.
Routing is opaque.
You do not get a clean receipt that says which model wrote which file.
For a warehouse, payroll or anything a client will audit, I would pin a known model for the production pass so we can reproduce the change.
I would also watch for cheap model drift.
A lower cost model can make a tidy looking change that is architecturally wrong, and you pay for that in the next week of fixes, not on the usage page.
What to do this week
Open a Replit project you already know.
In Agent settings, choose Power or Max, set the primary model to Automatically choose, and run one real task you could judge by eye.
Then open Settings, then Usage, and look at what that run actually cost.
If the result is as good as your usual pinned model, leave Auto on for iteration this week.
When you are ready to ship, pin the model for that pass so you can say what wrote the code.
If you want a second pair of eyes on how your team should set Agent modes, model policy and spend alerts, get in touch with Adevious AI.
Sources: https://replit.com/blog/intelligent-model-routing https://docs.replit.com/updates/2026/08/28/changelog https://docs.replit.com/chat/auto-mode https://docs.replit.com/features/agent/agent-modes https://thenewstack.io/replit-model-routing-default/
Acronyms: AI: Artificial Intelligence UK: United Kingdom