OpenAI is quietly testing a stripped-down version of ChatGPT that runs in a web browser without requiring users to log in. The company confirmed the experiment, which aims to cut inference costs by more than half compared to the standard logged-in experience.
What the lighter version does
The new app is designed for casual, one-off queries. Users land on a simple chat interface, type a question, and get an answer — no account creation, no history, no saved conversations. The model behind it is a smaller, cheaper variant of GPT, optimized for short interactions. OpenAI says the cost reduction comes from using a less computationally intensive model and from not having to store or retrieve user data.
Why OpenAI is trying this
The move comes as the company looks for ways to make AI more accessible while managing its own expenses. Running large language models at scale is expensive, and a lightweight option could let OpenAI serve more users without a proportional increase in server costs. It also lowers the barrier for people who are curious about ChatGPT but don't want to create an account.
What's not included
The logged-out version has clear limits. It cannot remember context across sessions, it doesn't support file uploads or image generation, and it won't show the model's reasoning steps. OpenAI has also restricted the types of queries it will handle — anything that could require sensitive data or prolonged back-and-forth is blocked. The company is testing these guardrails to see how well they prevent misuse.
How the test is rolling out
So far, the lightweight app is available only to a small percentage of users in select regions. OpenAI has not said when or if it will expand the test. The company is likely watching two things: whether the cost savings hold up under real traffic, and whether the reduced feature set leads to more abuse or user frustration.
The test is separate from the paid ChatGPT Plus and Team tiers, which continue to offer full access. It also doesn't affect the free tier for logged-in users, which still uses the standard model.
OpenAI declined to share specific numbers beyond the 50% cost reduction figure. The company also would not say how long the test will run or what criteria it will use to decide on a wider launch.
