Replit’s new default: Auto mode picks the best model for each task
The New Stack Paul Sawers
Replit is making its Auto model router the default for everyone. It’ll pick the model for each coding task, which is cheaper for Replit and less work for users.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Replit is putting its model-routing system at the center of its product. The AI coding company says Auto mode will become the default on every account, and the system will choose which model handles a task as that task changes. The pitch is simple: let Replit balance quality, speed and cost without making users think about it first.
That move lands right in the middle of a broader rush toward routing. Stripe agreed to buy OpenRouter for a reported $8 billion earlier in August. Ramp launched Router.com the same day, aiming to send requests to the cheapest model that still clears a performance bar. Cursor rolled out its own router in July, and Meta is reportedly working on an internal system called Switchboard that sends easier coding jobs to cheaper models.
Replit’s case for this is economics, plain and simple. Michele Catasta, Replit’s president and head of AI, says per-token prices can vary by orders of magnitude within a single model family, while cheaper models are getting close enough to the best ones for many tasks to leave room for savings. Replit says it spent months testing early versions of Auto mode, subagent routing and multiple iterations before shipping the public version.
The company already previewed the direction last week with Free Mode, a lower-cost Agent mode that uses Auto to pick a model and does not burn usage credits. Now that same setup becomes the default starting point for all users. Free Mode still has limits that reset every five hours, and users in that mode cannot manually choose a model. Core and Pro subscribers can switch to Power or Max modes if they want to turn Auto off and pick models themselves.
Enterprise customers get a different kind of control. Administrators can limit Auto to an approved set of models for each workspace, so Replit can keep routing automatically while staying inside company policy. Replit’s broader bet is that owning the agent and the surrounding infrastructure matters more than owning every model underneath it. Catasta says the company can change models as a task unfolds, and that its own usage data gives it signals a third-party router would not have.
My take — AI-written commentary, not fact-checked reporting
This is the sensible move, and also a very Replit move: hide the plumbing until the user actually needs a wrench. The bigger story is that model routing is turning into the boring but important layer of AI products, right where the hype usually hates to live. Fancy model names are nice; paying less for the same job is nicer.
Read more about this at: The New Stack
Related stories
Exclusive: Replit taps OpenAI's low-cost Luna model for new 'Free Mode'
Fortune ·
44
Auto Mode will soon be the default in Claude Code — because humans can’t be trusted
The New Stack · 3 weeks ago ·
20