The models inside Navōn, how Auto chooses, and comparing two at once

One of the quieter decisions in Navōn is which model answers you. There are thirteen of them in the picker, from three makers, and they are genuinely different: some write beautifully, some reason carefully through a hard problem, some are simply fast. This is a short guide to what they are, how Navōn chooses for you, and how to compare two of them when you want to see the difference yourself.

First, the part that does not change with the model. Every question is checked against your home's or school's standard before any model sees it, and every answer is checked again before you see it. Changing the model changes how an answer is written, never what is allowed.

The models, and what gold, silver and bronze mean

Each maker offers a line-up, from its most capable model to its quickest. In the picker their logos wear a small metal: gold for a maker's best, silver for its balanced middle, bronze for its fastest and lightest. The ranking is within a maker, not across them; gold on GPT-5.6 Terra means it is OpenAI's top model, not that it beats Claude Opus.

• Anthropic: Claude Opus 5.5 for long, careful writing and nuanced reasoning; Claude Sonnet 5 for everyday writing and explanation; Claude Haiku 4.5 when speed matters most.
• OpenAI: GPT-5.6 Terra for maths, logic and code that has to be right; GPT-5.6 Sol as a reliable middle choice; GPT-5.6 Luna for short factual answers and quick edits.
• Google: Gemini 3 Pro for very long documents, images, and questions that need the web; Gemini 3.7 Flash for fast everyday questions, also with the web; Gemini 3 Flash Lite for the simplest questions.
• xAI's Grok models are available only to adults on their own unsupervised accounts, and Auto never chooses them.

How Auto chooses

Most people should not have to know any of the above, which is why Auto is the default. When you ask a question on Auto, a small, very fast model reads it first, in roughly a third of a second, and decides three things: which model will answer it best, whether the answer depends on today, and whether you are actually asking for a picture or a video rather than words.

• It chooses from six models: Claude Opus 5.5, Claude Sonnet 5, GPT-5.6 Terra, GPT-5.6 Luna, Gemini 3 Pro and Gemini 3.7 Flash.
• When the answer depends on when you are asking, such as this week's parsha, candle-lighting, the news or a price, it turns on web search, and the answer is written by a Gemini model, which can search, with its sources linked underneath.
• When you ask it to make a picture or a video, it hands the request to the Image or Video tool, where the same modesty rules and image checks apply.
• If it cannot decide in time, the question goes to Claude Sonnet 5 rather than failing.

Above every answer on Auto you can see which model answered, with its logo, and a few words on why it was chosen. Nothing about the choice is hidden from you, and you can always pick a model yourself.

Comparing two models side by side

Sometimes the most useful thing is to see two answers next to each other: a teacher checking which explanation a class will follow, or anyone who wants a second opinion. In the model picker, choose your first model as usual, then press the compare icon on a second one. Send your question and both answers arrive side by side, each labelled with its model.

• Each answer has its own buttons: copy it, try it again with that model, read it aloud, share it, or report it.
• Try again never throws an answer away. Every attempt is kept as a version, and the arrows under the answer flip back through them, even when they came from different models.
• The same compare control works in the Image and Video tools, so you can see how two image models draw the same description.
• Only the first answer is carried forward as context for your next question, so the conversation stays coherent.

Making it yours

If you find yourself choosing the same model every time, set it once in Settings › Preferences: a starting model for Text, Image and Video, and the voice that reads answers aloud. Each tool opens on your choice, and you can still change it in the moment.

Whichever model you use, two things stay the same. The filter is the same filter, and no model is a posek: for a practical halachic question, the answer is to ask your rav.