How We Priced Thezis by Measuring Real Student Work

Why we stopped counting words

Words are easy to count, which is why many writing tools sell them. But the expensive part of a cited paper is not typing. It is finding sources, reading them, and checking that each sentence says only what its source says. A question about your argument writes no words at all and is still useful. A word limit charges for the cheap part and ignores the rest.

So Thezis Plus has two meters. Assistant usage covers everything the assistant reads, plans and writes. Searches count each look-up in published research. Both reset every month, and if you run out before a deadline, Extra usage tops you up and never expires. Pro is unlimited, with a fair-use limit of three times Plus. Free is a trial: 10 messages, with the assistant writing up to 1,000 words in each document. Nothing charges you automatically unless you turn that on. Thezis plans and pricing

We first ran this test on cheaper plans: a Free plan that reset every day, Plus at $10 and Pro at $20. It showed that daily Free covered most students’ whole month, which is generous but would not keep Thezis running. So Free became a trial, and the paid plans were priced closer to other writing tools. The measurements below are the same; the coverage and prices are for the current plans.

How we tested

We built a test that drives the real assistant: the same instructions, model, tools, research search and citation checks as the app. Only the student is scripted. Each scripted student sends the kind of messages a real student sends, and every search and every reply was real and paid for.

• Nine jobs, from asking questions about a paper to a work session on a 100,000-word dissertation. Each ran twice.
• Nine kinds of student, each a month built from those jobs, such as a busy undergraduate or a master’s student working on a thesis.
• Four runs under the real plan limits, where the student either waits for the limit to reset or buys Extra usage, the way a real student would.
• A session counted only if it left writing in the document. A session that ended without writing anything counted as a failure, not as a cheap session.

What each kind of work uses

Questions and polishing barely register. Searches are what run out first on paid plans, and adding sources to your own draft is the most search-heavy job we measured. Citation checks add up quickly because every cited sentence is checked against the passage it quotes: the paper with redos ran 89 of them.

Scaled to a full 1,500 words, a paper written section by section with every part redone once took about 23 messages and 17 searches. Redoing everything is expensive: it made the same paper use 1.4 times as much in one round of testing and 2.3 times as much in the next.

What each plan covers

Measured in that heavier kind of paper, the 1,500-word one with every part redone once:

• Free: enough to try Thezis on a short essay, not to write a full paper. The paper takes about 23 messages, and the assistant writes up to 1,000 words in each document on Free, counting text you reject or ask it to redo.
• Plus: about 10 papers a month. Searches run out first; assistant usage alone would cover about 15.
• Pro: unlimited, with a fair-use limit of about 31 papers a month, far more than any month we measured.
• Extra usage on Plus: $5 covers about 3 more papers, whenever you need them.

We also ran students into the real limits. A Plus student with three searches left at the end of a busy month bought $5 of Extra usage and finished the paper with almost all of it left.

What nine kinds of students pay

We built a month for nine kinds of student from those jobs. Their habits are our assumptions, not observations of real students. For each, we measured how much of a month of Plus their work would use.

Every month we built fits in Plus, including 16 long sessions on a dissertation. None needed Extra usage or Pro. Pro is for students who would rather not think about limits at all. Free’s 10 messages are enough to see how Thezis works on your own paper before you pay.

How that compares with other tools

We compared each month with the published plans of Jenni, Paperpal, Yomu and Samwell, as their pricing pages showed them on November 8, 2026. We did not run their products. We counted each Thezis message as one of their actions, which favors them, and read any limit with no stated period as monthly.

Plus costs about the same as the other tools’ cheapest paid plans. Paid yearly, it ties Yomu and costs less than the rest. Paid monthly, it costs the same as Paperpal, more than Yomu and less than Samwell and Jenni. For unlimited use, Thezis Pro at $50 a month costs less than Jenni Pro at $73 and Paperpal Pro at $59, and more than Yomu Ultra at $35.

Free plans differ more. Thezis Free is a one-time trial of 10 messages. Jenni’s free plan offers 5 chat messages and 3 edits. Paperpal’s allows 10 requests a day, which by its published limits would cover all nine months.

Counted by messages alone, Plus allows fewer per dollar than these plans: about 10 messages a month per dollar paying monthly, or 24 paying yearly, against 33 to 48 for the paid plans that publish a cap. A Thezis message in a paper often includes searching published research and checking each citation against its source, while an action elsewhere may be a single rewrite. We think the price of getting a month’s work done is the fairer comparison, which is why we led with it.

What the test found: citations typed as plain text

Running real sessions also showed us problems that no pricing spreadsheet would. The first was in the citations themselves. In our first runs of the 1,500-word paper, four of six sessions had at least one section whose sources were never linked. The assistant had found the sources, but it typed the citation as text instead of linking it.

Thezis now checks everything the assistant writes for sources named without a link. If it finds one, the writing goes back to the assistant to cite the source properly before it reaches your document. Citations you typed yourself are left alone. The assistant’s instructions now also say that linked citations are the citations, which the editor formats in your chosen style. In the runs after the fix, every section was linked, with 12 to 14 sources cited per paper and none typed as text.

What the test found: replies that stopped before writing

Each reply has a fixed amount of room, and the assistant’s planning counts against it. On long requests, such as several sections at once, it sometimes planned until the room ran out and wrote nothing. In an early test of the 3,000-word paper, three of five sessions ended with nothing written.

We replayed one of those requests four times at each setting. With the amount of planning we used then, one reply in four wrote anything. With less planning, four in four did. The assistant now plans less, sketches a long piece in a few words, and writes the first section right away. If a reply still stops with nothing to show, you see “The reply stopped before it finished. Try again.” instead of an empty reply.

In the final measurement, none of the assistant’s 899 steps ran out of room, and every session left writing in the document. Two of 925 had run out the time before.

What the test found: a usage meter that missed steps

When the assistant searches before it writes, one reply takes several steps: search, read the results, then write. Our meter recorded only the last step, so the reading in between went uncounted, about 4% of the real total in our test. Searches were always counted. The meter now records every step, and the plan coverage above was measured that way.

What this test does not show

• The students are scripted. Their habits are our assumptions, and real students write less predictably.
• Each job ran twice. Small samples move: the extra usage from redoing a paper measured 1.4 times in one round and 2.3 times in the next.
• We compared other tools’ published limits, read in their favor, as of November 8, 2026. We did not run their products, and prices change.
• We checked that the assistant delivered cited writing. We did not grade how good the writing was.
• If the assistant still types a citation as text after being sent back twice, the writing goes through rather than leaving you with nothing, so it is still worth a glance at your citations.

We will run the test again when the model, the limits or the prices change, and update this post.

Making your usage go further

• Ask questions freely. Asking about your argument, your sources or your structure uses very little.
• Polishing your own writing uses no searches at all.
• Redo a paragraph rather than a whole section. Redoing every part of a paper used up to twice as much.
• Ask for one section at a time. You can steer each part, and shorter requests are less likely to stop early.
• Watch the meter. On Free, Thezis shows how many of your 10 messages you have used. On Plus, it shows how much of this month’s usage you have used and when it resets.

Further reading

• Thezis plans and pricing
• Jenni pricing
• Paperpal pricing
• Yomu pricing
• Samwell pricing