Choose a model

When Floatboat works on something for you, an AI model does the thinking behind the scenes. Models are a bit like different helpers. Some are quick and easy on your credits. Others are better at hard problems and can take in more at once, but they use more credits.

Most of the time you don't need to think about it. The default, Lite mode, handles everyday work well. When a job is big or tricky, switch to a stronger model.

Switch the model for a taskCopy Anchor Link

A few things worth knowing:

• You can switch in the middle of a task. The round that's already running finishes on the old model. The new one takes over from your next message.
• Each task remembers its own model. Go back to an older task and the button still shows the model you picked there.
• New tasks start with the model you picked most recently. If you switched to a pricier model for one big job, switch back to Lite mode for everyday work. It saves a good number of credits.

How to pick a modelCopy Anchor Link

Each row shows, from left to right: the model's icon, its rate (like 0.1x or 1.5x), a bar, and the model's name.

• The rate roughly shows how fast a model uses credits. The bigger the number, the more credits the same job usually takes. The bar shows the same thing at a glance: the longer it is, the more it costs. The menu is sorted by rate, cheapest at the top.
• Lite mode always sits at the top. It isn't tied to one model. Floatboat picks a suitable one for you, which is why its note says smart routing and low-cost quick responses. For most everyday work, it's all you need.
• Good for appears when you rest the pointer on a model. Complex reasoning suits work that takes several steps of thinking or careful comparison. Long context suits reading a lot of long material in one go. Multimodal input means it can look at images directly.

Here's an easy rule: start with the default, and move up only if the result isn't good enough.

Lite mode is usually plenty for:

• Alex writing the weekly sales report from this week's orders.
• A student turning a research paper into five key points.
• A marketer drafting three social posts for a product launch.

A stronger model is worth it for:

• Alex going through all 13 weeks of Q3 orders to find what customers reorder most. That's the example in the video above.
• A teacher planning a full semester of lessons from the 12 units of a textbook.
• A product manager turning 30 customer interview notes into a feature list.

Not sure? Try the cheaper option first. If you're not happy with the result, switch to a stronger model and ask it to do the job again. There's no need to start with the most expensive one.

Long tasks: context compactionCopy Anchor Link

At every step, Floatboat looks back over everything said and read earlier in the task, so it knows what to do next. That's called the context. The longer the task, the more there is to look over. Replies get slower, they use more credits, and past a point the model can't hold it all.

So when a task gets long enough, Floatboat compacts the context on its own. It condenses the earlier part into a summary that keeps your goals, the key decisions, what's done, what's still open, and where the important files are. Then it carries on.

In the chat you'll see a line with Auto-compacting conversation context in the middle. When it's done, it changes to Conversation context auto-compacted. There's nothing you need to do. Floatboat keeps working.

Compaction doesn't delete your chat. Scroll up and every earlier message is still there. Only the material Floatboat carries forward to keep working gets condensed.

On a very long task, though, a small detail can get lost in the summary. If Floatboat seems to have forgotten something you asked for earlier, mention it again in your next message.

Can I change it?Copy Anchor Link

You can change when compaction starts. You can't turn it off.

A token is how AI measures the length of text. The longer the conversation and the more files it reads, the more tokens there are. The number you pick is how many tokens the context can reach before compaction starts.

• 500,000 (recommended) is the default and the largest option. Compaction comes as late as possible, so long tasks get interrupted the least and earlier details stay intact the longest.
• A smaller number means compaction comes sooner. Each later step has less to read, so it's usually faster and uses fewer credits. The trade-off is that earlier details get condensed sooner.
• Whatever you choose, Floatboat also compacts early when a conversation is about to outgrow what the current model can handle, or when going longer would bump it into a higher price per token.

If you're not sure, keep the recommended setting.

When a model is briefly unavailableCopy Anchor Link

Now and then, the model you picked can't be reached for a while. Instead of leaving your task stuck, Floatboat finishes that round with another model.

Afterward, move the pointer over the reply. In the row of small buttons that appears under it, there's an extra icon: an exclamation mark in a circle. Point at it to see which model actually did the work this time.

Your choice doesn't change. The button in the chat box still shows your model, and your next message goes to it first. If you think this round's result could be better, wait a little, then click Regenerate under the reply to have it done again.

Having trouble?Copy Anchor Link

Some models in the menu have a lock. Those need a paid plan. Rest the pointer on one and you'll see Upgrade to a paid plan to use this model. Clicking it opens the plans page in your browser. The Unlock more models card in the menu goes to the same place. For credits and plans, see Your account and credits.

A model you used before is gone. The list is updated online, and models can be retired or renamed. Pick one with a similar rate and Good for.

You switched models, but nothing seems different. The round that's already running stays on the old model. The new one starts with your next message.

Your credits are going faster than you expected. Check the model button in the chat box. You may still have a high-rate model selected, and new tasks use the model you picked last. Switch back to Lite mode for everyday work.

You see "Conversation context auto-compaction took too long, so this request was stopped." That round stopped, but your task and chat are still there. Send your last message again.

You see "Conversation context auto-compaction was interrupted." The compaction didn't finish. Maybe you clicked stop, or something went wrong along the way. Keep going by sending your next message. Floatboat will compact again when it needs to.

What's nextCopy Anchor Link