GPT-5.6 Sol is the best all-round maths model in Araveil: it is OpenAI’s flagship for maths, science and complex reasoning. For frontier-level problems, GPT-5.5 Pro uses maximum compute, at a much higher price. For step-by-step working at a low cost, Grok 4.20 (Reasoning) is on the Free plan.
For the hardest problems
Uses maximum compute for frontier problems. It costs far more, so keep it for problems that need it.
For problems in a photo
Reads images, so it can work from a photo of a worksheet or a handwritten problem.
Prices are each provider’s list price.
| Model | Plan | Input price | Output price | Reads up to | Speed |
|---|---|---|---|---|---|
| GPT-5.6 Sol | Lite | $4 per 1M tokens | $20 per 1M tokens | 1.1M tokens | Powerful |
| GPT-5.5 Pro | Pro | $30 per 1M tokens | $180 per 1M tokens | 1.1M tokens | Powerful |
| Grok 4.20 (Reasoning) | Free | $1.25 per 1M tokens | $2.50 per 1M tokens | 1M tokens | Powerful |
| Gemini 3.1 Pro Preview | Lite | $2 per 1M tokens | $12 per 1M tokens | 1M tokens | Powerful |
Maths punishes guessing. The models that do well reason before they answer: they work through a problem in steps rather than predicting the answer in one go, which is slower but far more reliable.
Showing the working matters as much as the answer, because it lets you see exactly where a solution went wrong, yours or the model’s.
For students, reading the problem as written matters too. A model that understands a photo of a worksheet saves retyping equations, and the mistakes that come with it.
Reasoning models are far more reliable than earlier ones, but they can still slip on long problems. Ask for the working and check it, especially before an exam or a report.
An Araveil setting, on the Pro plan, that gives a model extra time to reason before it answers. It helps most on problems with many steps.