T3 Chat FAQ

Can I trust that the AI is telling me the truth?

To put it simply: no. Large language models have gotten incredibly good, but they're far from perfect. It's good to double check all answers you get. Turning on search can help get more recent answers, but it's not a guarantee for correctness.

Why did T3 Chat give me an answer that's not related to my question?

AI models sometimes "hallucinate" and produce incorrect/irrelevant answers. Sometimes a simple "retry" will produce a better answer. As mentioned above, we'd recommend checking whether responses from the model are accurate or not, and not relying on them as a sole source of information.

Do you use my data for training?

We do not train our own AI models and have opted out of training data collection with our model providers where possible. Models on the following providers have been configured to not allow training on user inputs with the associated settings:

• OpenAI: Training is disabled by default, we have not opted in. Reference
• Anthropic: Training is disabled by default, we have not opted in. Reference
• Google: Training is disabled on paid models. Reference
• OpenRouter: We have explicitly disabled the option to route requests to providers that allow training on user inputs. Reference

Sometimes there are new models that can only be used with training enabled. When this is the case, we clearly indicate it so you can make an informed decision about whether to use that model.

If you are bringing your own API key, you are responsible for ensuring that your organization has configured their settings to not allow training if you do not want to allow it.

How does the usage meter work? It seems like it jumps around or drains faster sometimes.

The meter uses two buckets: a Base bucket and an Overage bucket. Usage always spends from Base first, then Overage. Base refills every 4 hours, and Overage refills on your monthly renewal date. If both are depleted, you need to wait for capacity to refill. In settings, these are shown as separate percentage bars so you can see each bucket independently.

Using simple "points" as an example, Base has 20 points and Overage has 200 points.

Whenever you send a message, we reserve an expected cost from your available balances. We apply it to Base first, then Overage. Once the response completes, we settle the final cost by either crediting back unused reserved amount or deducting any additional usage.

Because of this reserve-then-settle flow, your usage can temporarily move up before settling back down, which is why the meter may appear to go up and down at times.

Common things that can make cost per message go up:

• Long threads (more context is sent each turn)
• Large or many attachments
• Search-enabled requests
• More expensive models

Do you have feature XYZ?

We are always looking to improve T3 Chat, and are open to feature requests. If you have a feature you'd like to see, please let us know by opening a feature request on https://feedback.t3.chat.

How do I get support?

If you have any questions or need help, please contact us at support@t3.chat.

How do I set up custom search URLs?