You Shouldn't Have to Choose an AI Model. So We Don't Make You.

Open almost any AI assistant today and one of the first things it asks you to do is pick a model. A dropdown of names and version numbers: this one's faster, that one "thinks," another is "pro," a couple are labeled "mini." Most people have no idea which to choose, and they shouldn't have to. You came to ask a question, not to study a menu.

Ask Safely doesn't make you choose. Since August 2025, the app has picked the right model for each question on its own, and it tells you which one it used. That single decision is also why we can offer unlimited chat for free, with no message counter and no surprise downgrades. Here's how it works, and why the biggest labs are having such a hard time doing the same.

The dropdown was never built for you

Model pickers didn't start as a consumer feature. They came from the early days, when the people using AI were developers and researchers who wanted fine control and got used to running the most powerful model for everything. The dropdown made sense for them. It stuck around, and every new release added another option to it.

For everyone else, that menu is mostly friction. Choose the heavy model for a simple question and you wait longer than you needed to. Choose a light one for something that matters and you might get a weaker answer. Either way, you're being asked to make a call you don't have the information to make.

The bind the big labs are in showed up in plain view in August 2025. When one of the largest assistants launched its next generation, it tried to retire the picker and route every question automatically. Its longtime users revolted. They wanted their favorite model back, and they didn't trust a system choosing for them out of sight. The company restored the old models within days. Its own chief executive had called the picker a "confusing mess," and still had to bring it back.

That's the trap. These companies know automatic selection is better for most people. But they spent years training their power users to pick models, and those users won't give it up. So the complexity stays, and it grows with every launch.

We didn't inherit that habit

Ask Safely started as an app for regular people, not a tool for engineers. From day one, almost none of our users wanted to think about model names, so we made the choice for them and built the product around it.

It works simply. The moment you send a message, Ask Safely reads it for complexity. A quick, everyday question goes to one of Claude's faster models. A harder or higher-stakes question goes to a more capable one. You don't toggle anything. You ask, and the answer that comes back fits what you needed, in both quality and speed.

We show you which model answered

There's a fair objection to any system that picks for you. How do you know what you got? With most routers, you don't. The model is hidden, which is exactly why people have started asking other apps to put it back in view.

We show it. Each answer tells you which model handled it. If a question matters and you want to confirm it went to the stronger model, the information is right there. Routing you to a cheaper model in silence is a downgrade. Routing you to the right model and telling you which one is a feature. That difference is the whole point.

This is why the free tier is actually free

Here's the part that connects to your wallet. Running a top-tier model on every trivial question is expensive, and someone always pays for it, usually you, through tight limits. A faster model can handle a large share of everyday questions at a fraction of the cost. Independent research on AI model routing found you can keep about 95% of a top model's quality while sending only around a quarter of questions to the expensive one, which cut cost by roughly half.

Because we're not spending frontier compute on "what should I make for dinner," we can give everyone unlimited chat on the free tier. No daily message count. No "you've used your ten messages, you're switched to a weaker model now," which is how most free plans quietly work. You ask as much as you want.

The paid plan adds the extras: access to Opus, Claude's most capable model, plus voice conversations, document uploads, and deep research. But the core promise doesn't sit behind a paywall. Free means free, and unlimited means unlimited.

The version most people actually wanted

The model picker is a habit the industry built and now can't easily break. We never inherited it, so we don't have to defend it. You ask a question. We pick the model that fits, tell you which one we used, and never stop you at a limit. For most people, that was the point of an AI assistant all along.