What AI model does it use?

What AI model does Replit use?

Replit's coding agent used to be thought of as a Claude product. The docs I read on September 25, 2026 tell a more mixed story, and the default isn't Claude at all.

Free Mode: a router, launched on GPT-5.6 Luna

Replit Agent has three modes now: Free, Power and Max. (Economy Mode is gone; the docs say "Power Mode replaces it at the same price.") Free Mode is the one most people land in, and you can't choose its model. "Free Mode always uses intelligent model routing and cannot select a model manually," the Auto mode page says.

When Replit announced Free Mode on August 18, 2026, Amjad Masad and Michele Catasta wrote that it was "powered by OpenAI's GPT-5.6 Luna", in a post subtitled "With OpenAI, Replit now helps you accomplish even more." Eight days later, on August 26, Replit made its model router available to everyone, and Free Mode has used routing ever since.

That leaves a gap I couldn't close. The current docs name no model for Free Mode, and Replit doesn't list the models its router chooses between, so whether Free Mode still runs on GPT-5.6 Luna alone isn't something Replit has said.

Power and Max: the models you can pick

On the Core and Pro plans you can choose a model by hand, from a picker inside Power or Max Mode. The model selector page listed these on September 25:

Replit adds that "the list in your Model selector is the source of truth for your account", because model selection is still rolling out. Effort (Low up to Max) is a setting on each model, not a mode of its own.

What your plan changes

Starter users don't get the picker. The Free Mode page says "Builders on Starter need to upgrade before using Power Mode or Max Mode." So on Starter, the only honest answer to "which model?" is whatever the router sends you to.

Enterprise admins get a different control: they can turn whole providers on or off (OpenAI, Anthropic, Google, DeepSeek, Kimi and Z.ai are on the list), but not all the way down to one. "At least one model from each of OpenAI, Anthropic, and Google must remain enabled," the docs say, so a company can't run the agent on a single lab.

Why the router matters

Replit's case for routing is cost. In its August 26 post Replit claims "the same output quality at 65% lower cost than the previous" Max Mode. That's Replit's own figure. It also means the model answering any given request in Free Mode is Replit's choice, made per request, and the product doesn't promise to tell you which one it was.