Claude for students: free tier, university, API

What does Claude cost a student?

For most students, Claude costs either nothing or the full retail price, because as of October 2026 there is no individual student discount in between. Anthropic's published pricing carries no verified-student rate and no .edu code. The one discounted academic tier that does not need a university-wide deal is the Claude Team plan for scientists, and a student cannot buy it alone: a principal investigator applies for the research group and can then add graduate students to it. What most students get is a free plan that covers a lot of coursework, a university-wide plan that the institution buys, and a pay-as-you-go API that can run a whole term for the price of a coffee.

This post states exactly what is offered and what is not, with every price stamped with the date I read it. Prices move. Treat claude.com/pricing as the source of truth and this page as the explanation.

Is there a student discount on Claude?

No, not as an individual purchase. Here is the precise shape of it as of October 2026.

• Not offered: a discounted Pro, Max or Ultra price for verified students, a student promo code, or a cheaper rate for a .edu email address. None of these appear on the pricing page.
• Offered: Claude for Education, described by Anthropic as "a comprehensive university-wide plan for an institution, including its students, faculty, and staff". The institution buys it. A student cannot purchase it directly, and there is nothing to apply for either: students use it through a university-sponsored account by signing in with their university email.
• Offered, if you work in a research group: the Claude Team plan for scientists, read on 21 October 2026. A principal investigator or equivalent at an accredited university or nonprofit research institute, working in the natural sciences, mathematics, computer science, engineering or a related field, applies and verifies their affiliation, with a minimum of 1 seat and up to 25 per group. Once approved, Anthropic's help center says they can add "postdocs, graduate students, and researchers". Standard seats are $0 per user per month and Premium seats are $15 per user per month, or $180 a year on annual billing, against regular prices of $20 and $100. The promotional pricing lasts 12 months from sign-up, after which seats renew at the then-current price. Student verification alone does not qualify you, and the help center does not mention undergraduates.
• Not a discount either: Claude Campus, Anthropic's student program, pays selected student leads a US$3,600 stipend and gives Builder Club members resources such as API credits. It is a role you apply for, not a price. Its page says the program is closed for Fall 2026, and applications ran from 1 to 12 September.
• Also true: the free plan is real and is not a trial.

Any site selling you a "Claude student code" is selling you something Anthropic does not issue. Some of those pages resell shared logins, which breaks the terms you agreed to and can get the account closed. There is no code to hunt for, so stop hunting and pick one of the three real routes below.

The retail prices those routes are measured against, read on 31 October 2026: Free is $0. Pro is $20 per month billed monthly, or $17 per month on the annual plan, which bills $200 up front. Max is $100 per month and Ultra is $200 per month, both billed monthly on the web. Team seats start at $20 per seat per month billed annually. If you are weighing the tiers against each other, which Claude plan fits the way you work walks through the whole lineup.

What the free plan actually gives you

The free tier is the first thing a student should exhaust, because it is more capable than its price suggests. As of October 2026 it includes web search, file creation, code execution, connectors, and memory. It gives you the Sonnet and Haiku models; Opus and Fable are reserved for paid plans. For most coursework that is enough, and what you are mainly buying on a paid plan is volume and plumbing, plus access to those top models. Most students are choosing between this free tier and ChatGPT's rather than between two paid plans, and the two ladders priced side by side covers what each vendor gives away for free as well as what the first paid step costs.

Two things the free plan does not include matter to students. Claude Code, the terminal coding agent, starts at Pro. And the free allowance is small enough that a heavy afternoon of essay drafting will hit it. Before you decide the upgrade is worth a student budget, it is worth reading where Pro's own usage limits stop you, because paying does not remove the wall so much as move it further out.

When you hit it you get a message naming the reset time, something like You've hit your session limit. That is a subscription limit, not an error, and switching model does not clear it. What to do when you hit a Claude usage limit covers the difference between that wall and the API's separate rate limits.

For a student on free, the practical rhythm is simple. Do the thinking-heavy work in one focused block, because the session allowance resets five hours after your first message rather than at midnight. Save the long back-and-forth for when the window has refilled. Free also has a weekly limit, so a session reset does not always restore access. If that rhythm stops working and the wall turns up most weeks rather than most terms, whether Pro earns its $20 comes down to counting how often the free limit actually stops you.

Claude for Education: the route through your university

Claude for Education is the only official discounted access open to every student, and your institution has to be the buyer. If your university has signed up, you get access by signing in with your university email, at no personal cost. Ask your IT service desk or the library first, since the licence often exists before students hear about it. If the answer that comes back is that your department buys seats on a business contract instead, what a Claude Enterprise seat actually costs explains why the advertised seat fee is only part of that bill.

The version sold to universities adds a study-oriented behaviour Anthropic calls learning mode, which it describes as working "like a good tutor: it asks questions that help you find the answers yourself" so that students build transferable skills instead of collecting answers. That framing matters for academic integrity policies, which is usually the reason a department agrees to it.

If you are a graduate student in a research group, there is a second route that needs no university-wide contract. Your principal investigator can apply for the Claude Team plan for scientists and add you to a seat. As of 21 October 2026 a Standard seat is $0 and includes Claude Code, and a Premium seat with five times the usage is $15 a month, both for the first 12 months. The plan also includes monthly API credits of $20 per Standard seat and $100 per Premium seat, pooled across the group up to $500; a workspace owner claims them by linking a Console organization after seven days on the plan. You cannot apply for it yourself, so that request goes to your PI rather than to IT.

If your university has not signed up, one email to whoever owns software licensing costs you nothing. If they have, you are done reading.

The API is the real budget path

If you want more than the free plan and $20 a month is not in your budget, the Claude API is the answer most students skip. You create an account on the Claude Console, buy prepaid usage credits, and pay per token from that balance with no monthly fee and no plan allowance to run out of. The API is not a second free plan, though: there is no free tier, only a small starter credit. Anthropic's pricing page says "New users receive a small amount of free credits to test the API", without naming an amount, so treat it as enough to try your first requests and check the balance in the Console rather than planning around it. Past that, Anthropic's billing help says "Buy credits before you use the API", and successful API calls and completed tasks draw from your credit balance unless your organization has a monthly invoicing arrangement. Failed requests are not charged. If your client disconnects or times out during a request that was on track to succeed, that request is still charged. Max, Ultra and Team plans automatically include monthly API credits with no selection or approval step; Free and Pro plans are not eligible. Max includes $100 a month in API credits and Ultra includes $200; you claim them by linking a Console organization after seven days on the plan, and unused credits expire each billing cycle. These monthly credits do not cover interactive Claude Code sessions or extra usage in the Claude apps. For students the realistic sources beyond the starter credit are Anthropic's own programs: Claude Builder Club members receive resources such as API credits, and Anthropic's Claude for Education announcement invites students who are building projects with Claude to apply for funding in the form of API credits. Both programs need you to be selected or approved, so do not plan a term around them. The API has rate and monthly spend limits, and requests stop when prepaid credits run out. For a coursework workload that meter is very cheap.

The prices per million tokens (written as MTok in the docs), as of October 2026:

• claude-haiku-4-5: $1 input, $5 output. 200K context.
• claude-sonnet-5: $2 input, $10 output. That launch rate is now the standard price; the rise to $3 and $15 that had been scheduled for 1 September 2026 was cancelled. 1M context.
• claude-opus-4-8: $5 input, $25 output. 1M context.

All three still work, but Anthropic now lists them as legacy models. The current lineup is Haiku 5.5, Sonnet 5.5 at $2 and $10, Opus 5.5 at $4 and $20, and Fable 5.1. The one that matters on a student budget is claude-haiku-5-5: $0.10 input and $0.50 output for prompts up to 100,000 tokens, $0.50 and $2.50 above that, with a 1M context. Haiku 4.5's earliest retirement date, not sooner than 15 October 2026, has now passed. On 21 October 2026 Anthropic's model deprecations page still listed it as active, and Anthropic promises at least 60 days' notice before a retirement, so check that page before you build a routine around it.

The word count represented by a million tokens depends on the model's tokenizer and the text. Sonnet 5, Opus 4.8 and the newer models use the tokenizer introduced with Opus 4.7, which counts the same text as roughly 30% more tokens than earlier models such as Haiku 4.5. Either way the unit is much larger than a student workload. That is the whole reason the API wins here.

The catch is honest and worth stating: the API does not include the Claude app's chat interface, though the Console has a playground for testing prompts. You get a key and an endpoint, and you either use a client that accepts a key or write a few lines of code. One request looks like this.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-haiku-4-5","max_tokens":1024,"messages":[{"role":"user","content":"Explain the Nyquist limit in two paragraphs."}]}'

A healthy response is JSON containing a content array with the answer and a usage block reporting input_tokens and output_tokens. Those two numbers are your bill for that request. If you get {"type":"error","error":{"type":"authentication_error"}} back instead, the key in ANTHROPIC_API_KEY is unset or wrong, because the API rejects the request before the model ever sees it.

If you would rather keep a chat window, the arithmetic behind the API and a flat plan shows where the two lines cross.

A worked month for a coursework workload

Numbers beat adjectives, so here is a full month costed out with the assumptions written down. Change the assumptions and the answer changes, which is the point.

Assume a study question sends about 2,000 input tokens, which is your question plus a page or two of lecture notes, and gets back about 800 tokens. On claude-haiku-4-5 that is 2,000 at $1 per MTok, which is $0.002, plus 800 at $5 per MTok, which is $0.004. Call it $0.006 per question.

Fifteen questions a day across 20 study days is 300 questions, or about $1.80 for the month. On claude-haiku-5-5 the same 300 questions come to about $0.18, or roughly $0.23 once you allow for its newer tokenizer counting around 30% more tokens for the same text.

Now add the heavier jobs. Summarising a 40-page reading is roughly 30,000 input tokens and 1,500 output tokens. On claude-sonnet-5 at its standard rate that is $0.06 of input plus $0.015 of output, so about $0.075 per reading. Ten of those in a month adds $0.75.

Total for the month: a little over $2.50, against $20 for Pro. The gap holds until your usage stops looking like coursework and starts looking like a job, which for most students is around the point where you are running an agent all day rather than asking questions. If a dissertation month or an exam term does push you onto Pro, stepping back down once the deadlines pass keeps the month you have already paid for, so a subscription can be a term-time expense rather than a standing one.

One more lever if you are patient. On Anthropic's Batch API, "All usage is charged at 50% of the standard API prices", input and output alike. Most batches finish inside an hour and anything not done in 24 hours expires. For a pile of readings you want summarised by tomorrow rather than right now, that halves an already small bill.

Pick the cheaper model on purpose

Model choice moves a student's bill more than anything else, because every model in the range prices output at five times its input rate. Haiku 4.5 costs a fifth of Opus 4.8 on both sides of the meter.

For mechanical work where you already know what a correct answer looks like, such as reformatting citations, extracting definitions, or translating notes, a Haiku model is the right default, and Haiku 5.5 is now the cheapest on the price list. Reach for Sonnet 5 when the task is genuinely open, like reasoning through a proof or reviewing an argument. Opus 4.8 earns its price on hard agentic work, which is rarely what an essay needs. Claude Fable 5 and its successor Fable 5.1 sit above all three at $10 input and $50 output per MTok, which is ten times Haiku 4.5 on both sides of the meter, so it is worth knowing what that premium buys before you let it near a coursework budget. Choosing between Opus, Sonnet and Haiku goes through the tradeoff properly.

Before you commit a big prompt, you can price its input side closely. The token counting endpoint is free and tells you the input size before you spend anything on output.

curl https://api.anthropic.com/v1/messages/count_tokens \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-haiku-4-5","messages":[{"role":"user","content":"Explain the Nyquist limit."}]}'

The reply is a single field, input_tokens. Multiply it by the input rate above and you know roughly what the input will cost before you send it. Anthropic's docs call that count an estimate, and the billed input can differ from it by a small amount. The output side cannot be counted in advance at all, because it depends on how long the answer runs, so cap it with max_tokens and budget for that ceiling. Do not estimate token counts with a tokenizer built for another vendor's models, because the counts differ enough to make your budget wrong. The same goes for reusing a count from an older Claude model: Opus 4.7 and later models count the same text as roughly 30% more tokens, so count against the model you will actually call.

Keep your numbers easy to refresh

Every figure on this page has a date on it because AI pricing changes several times a year, and a stale number is worse than no number. Two habits keep you accurate.

Read the pricing page on the day you decide, not the day you researched. The plan fees and the per-token rates both live there, and a post like this one is a snapshot by definition.

Then read your own meter rather than anyone's estimate. On a subscription the /usage view shows your remaining allowance. On the API, the usage block on every response is the ground truth for that request, and the Console shows the running total. Your own numbers are always more accurate than a worked example built on someone else's assumptions.

FAQ

Does Anthropic offer a student discount on Claude Pro?

No. As of October 2026 there is no individual student rate, student promo code, or academic discount on Claude Pro, Max or Ultra. The main official education pricing is Claude for Education, which is a university-wide plan that the institution buys rather than something a student can purchase. Graduate students can also get a $0 or $15 seat on a lab's Claude Team plan for scientists, but only after their principal investigator applies and is verified. Sites advertising Claude student codes are not issuing anything Anthropic recognises.

Is the free Claude plan enough for university coursework?

For most students, yes, with one caveat. The free plan as of October 2026 includes web search, file creation, code execution, connectors, and memory, with the Sonnet and Haiku models, so it handles reading, drafting and explanation work. It does not include Claude Code or the Opus and Fable models, and its usage allowance is small enough that a long drafting session will hit the limit. When it does, you wait for the five-hour session or weekly limit to reset, depending on which one you hit.

How do I get Claude for Education at my university?

You cannot buy it yourself. Ask your IT service desk, library, or whoever owns software licensing whether the institution has a Claude for Education agreement. If it does, your access comes through your student account at no personal cost. If it does not, the request has to come from a department rather than from an individual student.

Is the Claude API cheaper than a subscription for a student?

Usually, yes. A coursework workload of roughly 300 short questions plus ten long readings costs a little over $2.50 a month on the cheaper models at October 2026 rates, against $20 a month for Pro. The API charges per token with no monthly fee, so light use costs very little. It becomes more expensive than a plan once you are working with Claude for hours every day rather than asking questions.

What do I lose by using the API instead of a plan?

The Claude app's interface, mainly. The Console has a playground for testing prompts; for your own chat setup, you supply a client or write a small script. You also lose Claude Code's subscription allowance, though purchased API credits can pay for Claude Code, and the API supports both five-minute and one-hour prompt caching. In exchange you get exact per-request costs, subject to API rate and spend limits and your available credit balance.