Best AI for students

Students ask an AI model to explain a proof, check a derivation, summarise a 40-page PDF or turn a recorded lecture into notes, usually on a small budget. This ranking weighs evidence on reasoning against price and the features that coursework needs.

Who this page is for

This page is for students in school, university or self-study who want one AI model for homework help, maths and science problems, reading papers and slides, and working with lecture recordings. It assumes you use a chat app or a low-cost API key, so cost matters a lot.

In the app or via API? The price in this ranking is the API list price per token. If you use a chat app, our free AI plans compared page lists the free and paid consumer plans with their prices, included models and usage limits.

How this ranking works

The ranking combines six criteria into one score. The weights are fixed and add up to 100%. Our methodology explains how each criterion is scaled and combined.

• Maths reasoning, 35%. Epoch AI runs FrontierMath Tiers 1-3 (v2) itself on hard, unpublished problems, with the model allowed to use a Python tool ^[1]. Maths and science problems are the hardest part of most coursework, so this gets the largest weight.
• Chart and figure reasoning, 10%. CharXiv Reasoning asks questions about charts and figures from scientific papers, answered without tools. Coursework often comes with graphs and diagrams.
• Blended price, 25%. Students often pay out of pocket, so a low price matters. Lower is better, on a log scale, so that a few expensive models don't flatten the differences among the rest.
• Reasoning mode, 10%. Step-by-step reasoning helps with maths and science homework.
• Document and PDF input, 10%. Lecture notes and readings usually come as PDFs.
• Audio input, 10%. Audio input lets you work with recorded lectures.

Which benchmark results count

Each benchmark criterion uses a setting group: a list of stored settings we treat as comparable for this ranking.

• FrontierMath: every Tiers 1-3 (v2) run by Epoch AI with its Python-tool harness counts, at any reasoning effort, including runs without a thinking mode.
• CharXiv Reasoning: runs on the reasoning questions without tools or code interpreter count, at any effort, including runs where the effort is not stated.

Within each group, a model counts with its best result. Independent results are preferred over results a provider reports itself; if only a provider result exists, it is used and marked "self-reported". The methodology box on this page lists every stored setting in each group.

A model needs data for at least 60% of the total weight to be ranked. Price and the three capabilities together make up 55%, so a model needs at least one of the two benchmark results. A model with a CharXiv result but no FrontierMath result can still be ranked, without any maths evidence.

How to read the results

• Scores are relative. Each criterion is rescaled from 0 to 100 across the candidates on this page. 100 means best among these models, not perfect, and adding or removing a model can shift the others.
• Check the setting. The table shows the setting behind each score, including the reasoning effort.
• "Self-reported" marks a result that comes from the model's own provider rather than from an independent run.
• Small gaps are noise. Runs at different efforts sit in the same group, so treat models close in score as roughly equal.
• Look at the criteria that matter to you. If you never record lectures, audio input is worth less to you than its 10% weight. The table shows each criterion separately.

Limitations

• Grouped runs differ in reasoning effort. Because each model counts with its best run, a model tested at many effort levels has more chances to score well.
• OpenAI funded FrontierMath and has access to part of the problems ^[1].
• Many CharXiv results come from providers' own model cards ^[2]; the table marks them "self-reported".
• This ranking uses API list prices. Consumer subscriptions and free tiers, which decide what most students actually pay, are not part of it; AI subscriptions compares the app plans separately.
• Consumer apps can use different model versions, limits and tools than the API, so a model's API results may not match what you see in the app.
• Only models with a result in at least one benchmark group can be ranked. A missing result doesn't mean a model is weak, only that we have no comparable measurement.
• No benchmark measures whether an explanation helps you learn. Check AI answers against your course material, and follow your institution's rules on AI use.
• The data behind this page is as of October 2026.

How to choose for your own work

A short shortlist and your own test will beat any general list. Some catalog facts help narrow it down, as of October 2026:

• Budget. Claude Haiku 5.5 and GPT-6 Luna both cost $0.10 per million input tokens and $0.50 per million output tokens on the API ^{[3] [4]}. Gemini 3.5 Flash-Lite costs $0.30 and $2.50 ^[5]. The AI pricing page lists every model, and the cost calculator turns your usage into a monthly figure.
• Maths and step-by-step problems. Choose a model with a reasoning mode and compare FrontierMath results on the model pages. Reasoning models explained covers when the extra thinking is worth its cost.
• PDFs, notes and slides. Claude and OpenAI models accept PDF files directly ^{[6] [7]}. Most models read documents; check the model page before you rely on it. Context windows explained shows how much text fits in one request.
• Lecture recordings. Gemini models such as Gemini 3.8 Flash accept audio input ^[8], and so does Muse Spark 1.1 ^[9]. The consumer-available Claude and GPT-6 models do not.

Then test two or three candidates on the same real tasks: a problem set you can grade, a PDF you know well and a short recording. How to compare AI models explains how to do this fairly, and the comparison page puts the candidates side by side.

Related guides

• How to choose an AI model
• Reasoning models explained
• Multimodal AI explained
• AI token pricing explained
• Prompt engineering basics
• All guides