Replit Free Mode: Unlimited Light AI Work on Paid Plans

Replit has taken the meter off the part of AI building where most of your time actually goes.

What happened

On 19/08/2026 Replit launched Free Mode, powered by OpenAI's GPT-5.6 Luna model. The two companies made the announcement jointly and gave the story to Fortune first.

The name needs unpicking. Free Mode is not a free plan. It sits inside the existing paid tiers, which run at 20 US dollars a month for Core and 100 US dollars a month for Pro. What is free is the usage. Talking to the Agent, asking it questions about your project, planning a feature, sanity checking an approach or running something light no longer eats into your credit budget.

When a job genuinely needs more capability, the Agent routes it to the larger GPT-5.6 Sol model, gets the work done, then drops back into Free Mode while holding on to the full project context. Free Mode is now the default for Core and Pro subscribers. Michele Catasta, president and head of AI at Replit, calls that choice radical, and for a company that bills by consumption, it fairly is.

The economics behind it are plain enough. OpenAI cut the price of Luna by 80% on 30/07/2026 for developers using its API. Thibault Sottiaux at OpenAI told Fortune the reduction came from running the model more efficiently rather than from a sudden influx of new compute. Replit took that saving and handed it back as unmetered usage instead of a lower sticker price.

Why this matters for UK businesses

Anyone who has built on a credit based AI platform knows the feeling. You stop asking questions. You batch up your thinking so you do not waste a run. You skip the conversation that would have caught the flaw and go straight to building the wrong thing. That behaviour is expensive in a way no invoice ever shows.

For a small UK business, the planning conversation is usually where the value sits. Working out whether you need a full booking system or a shared calendar with a form in front of it is a five minute chat that saves a fortnight of wasted effort. Metering that chat was always the wrong incentive.

There is a second point worth noticing. Replit has leaned heavily on Anthropic models for some time. Putting an OpenAI model at the centre of its default experience, and saying openly that more joint launches are coming, tells you these platforms are becoming properly multi vendor. If you build on them, you are no longer betting on a single model provider.

How we see it at Adevious AI

We would switch Free Mode on and change how we use the Agent, not just how much we use it. The habit worth forming is to treat the Agent as a colleague you think out loud with before you ask it to build anything. Describe the problem, argue about the shape of the solution, ask it to poke holes in your plan. That part is now unmetered, so use it properly.

Two cautions. First, Luna is the small, low cost model in the GPT-5.6 family. Fast and cheap is the right trade for chat, triage and quick answers. It is not the model you want quietly making architectural decisions on your behalf, and the routing between Luna and Sol happens automatically rather than on your say so. Check what the Agent has actually decided before you let it build.

Second, unmetered exploration is not the same as unmetered building. Your build budget is unchanged, and a longer, richer planning conversation can easily produce a bigger and more expensive scope than you intended. Free thinking has a way of turning into paid work. Set the scope yourself and hold it.

We would also weigh the vendor concentration. Free Mode runs on one model from one provider. That is fine for planning, but if a supplier switch or a price move would hurt you, keep the critical logic in your build portable rather than shaped around one vendor's habits.

What to do this week

Pick one internal process that annoys you, the sort of thing that lives in a spreadsheet and three email threads. Open Replit, stay in Free Mode, and spend twenty minutes describing it to the Agent without asking it to build a single thing. Ask what the smallest useful version looks like, then ask what it would cost you to run and maintain. Only when you can answer both, move into Build Mode. You will spend fewer credits and end up with something you actually keep.

If you would like a hand working out what is worth automating in your business, and what is better left alone, get in touch with Adevious AI. We are always happy to talk it through.

Sources: https://openai.com/index/replit/ https://fortune.com/2026/08/19/exclusive-replit-taps-openais-low-cost-luna-model-for-new-free-mode-subscription-tier/ https://openai.com/index/gpt-5-6/ https://docs.replit.com/updates/2026/08/07/changelog

Acronyms: AI: Artificial Intelligence API: Application Programming Interface UK: United Kingdom