Updated on
SciSpace: reviews and analysis
Question your own PDFs and a paper pool the vendor declares at 280 million, from one screen.
Affiliate link · no extra cost to you
Our verdict
SciSpace bundles two things this segment usually keeps apart: questioning your own PDFs and searching a pool the vendor declares at 280 million papers. As a reader with an assistant beside you it delivers, and Capterra users give it 4.3 out of 5. The trouble is at the till. Everything runs on credits that expire at the end of each cycle, the pricing page itself opens on the annual rate, and what a task will spend is unknown until it finishes, because it depends on the computing power required. Anyone taking the annual plan should read the renewal terms first.
Best for: Anyone reading papers daily who wants to search and ask on one screen, knowing they are paying by credits.
What the internet says
SciSpace, formerly Typeset, does two jobs on the same screen: questioning the PDFs you upload and searching a pool of papers by meaning. The size of that pool needs care, because the party contradicting itself is the vendor: its home page and its search page say more than 280 million, while the pricing page says more than 270 million in its description and more than 200 million in the body. The 280 million figure is also the one Purdue University Libraries' guide carries, so that is the one published here, attributed rather than stated as hard fact. As a reader with an assistant beside you, third-party assessment is good: 4.3 out of 5 across 80 verified Capterra reviews, where the praise that repeats is the comfort of reading, summarising and chatting with papers, and the fact that it links out to the originals. The objections are barely about answer quality and almost entirely about the till. The complaint that repeats most on Capterra is credit consumption 'without the option to predict the final cost for a simple task', and in the vendor's defence it does publish an indicative table of scenarios and a real-time ledger, so the fair word is unpredictable per task rather than opaque. There are also complaints of renewal charged with no advance reminder. A fourteen-day hands-on review, from a site that sells its own tools and is therefore not disinterested, adds service outages especially at weekends and a search over your own files that stops responding past a certain letter, though it contradicts itself on refunds. The weakness in the hard sciences comes from one individual review by an engineering professor, not from a measured pattern. The price needs care: the page opens on the annual rate, where the entry plan reads $12 a month billed annually, while the same plan billed monthly is $20. The currency depends on where you are: from Spain and Latin America the site publishes dollars, and only in India does it switch to rupees.
What the web repeats in favour
- It gathers chat over your own PDFs and semantic search across its paper pool in one place
- It links to the real papers, so checking that a reference exists takes one click
- Eighty verified Capterra reviews give it 4.3 out of 5, which is measured adoption rather than impression
- It analyses folders of PDFs in batches rather than paper by paper
What the web repeats against
- What a task spends is unknown until it finishes, because it depends on the computing power it needs
- Monthly credits expire at the end of each cycle and do not roll over
- Several verified reviews complain about automatic renewal charged with no prior notice
- An engineering professor gives it one star on Capterra, saying it falls short in the hard sciences
- Someone who used it for fourteen straight days reports service outages, especially at weekends
- The pricing page opens on the annual rate, so the first figure you see is not what paying month to month costs
- The vendor itself publishes three different sizes for its paper pool across three pages of its own site
Sweep sources: Official pricing · Review sites · Review sites · Review sites
Pros / Cons
Pros
- Joins chat over your own PDFs with semantic search across the paper pool the vendor declares
- Capterra users rate it 4.3 out of 5 across 80 reviews
- It links to the real papers, so checking that they exist takes one click
Cons
- What a task spends is unknowable in advance, and credits expire at the end of each cycle
- Billing complaints repeat: renewal charged with no advance reminder unless you cancel
- An engineering professor gives it one star on Capterra, saying it falls short in the hard sciences
TLDR: SciSpace does two jobs on one screen, questioning the PDFs you upload and searching a large pool of scientific papers by meaning. As a reader with an assistant beside you it delivers, and its verified users rate it 4.3 out of 5. The trouble is at the till: everything runs on credits that expire each cycle, and what a task spends is unknown until it finishes, because it depends on the computing power required.
What SciSpace is and how it works
SciSpace used to be called Typeset and came out of the world of academic publishing templates. Today typeset.io redirects here and the product is something else: a research assistant that joins two capabilities this segment almost always keeps apart.
The first is the usual one in this category. You upload your PDFs, it reads them and you hold a conversation with them. You can do it with one, with a whole folder, or in batches, which is where you notice it was built for people who read a lot rather than for consulting a single document.
The second is what separates it from a pure document tool: it also searches a large pool of papers by meaning. The size needs careful wording, because the vendor contradicts itself. Its home page and its search page say more than 280 million papers, the pricing page says more than 270 million in one place and more than 200 million in another, and its data sources page loads empty. The 280 million figure is the one Purdue University Libraries’ guide carries, so that is the one given here, attributed rather than settled. Whatever the number, what you can do is ask what exists and bring it back. It links out to the originals, so checking that a paper is real takes one click, which is the difference between this and an assistant that invents well-written references.
That is also why it is not a hollow wrapper: there is an index of its own, semantic search over it and batch analysis, three things the underlying language model does not ship with.
What using it is like day to day
The natural workflow is a reader with an assistant alongside. You open the paper, and in a panel you can ask it to explain a formula, summarise a section or clear up a term. Capterra users, where it holds 80 verified reviews and an average of 4.3 out of 5, agree on two points of praise: it saves the raw reading of a pile of papers, and the interface stays out of the way.
That is where the unanimous good news stops. The part a buyer has to read in full is about consumption, because nearly every serious complaint lands there rather than on answer quality.
Everything is paid for in credits. Each action spends an amount you do not know in advance, because the vendor itself explains it depends on the computing power the query needs, and the complaint that repeats most across verified reviews lands exactly there: no way to predict the final cost of a simple task. In fairness it does publish a table of scenarios with indicative figures and a real-time ledger of credits issued, spent and remaining, so you are not flying blind. What does not change is that monthly credits do not roll over, they expire at the end of each cycle, so a quiet month is money that evaporates.
On top of that there is a pattern of billing complaints. Several verified reviews warn that renewal is charged with no advance reminder unless you cancel. A review from fourteen days of continuous use adds service outages with some frequency, especially at weekends, and a search over your own files that stops responding past a certain letter, which in a large library is as annoying as it sounds. That piece is worth reading with your guard up, because it comes from a site selling its own tools and because it contradicts itself: it says getting a refund is almost impossible and also lists the refund process among the things done well.
On the answers themselves, criticism is more spread out and worth attributing rather than generalising. The most quoted comes from an engineering professor who gives it one star on Capterra and writes that it falls short in the hard sciences, adding that in microfluidics it was a waste of time and that it might suit the social sciences better. That is one specific opinion, not a measurement. The person who tested it for fourteen days says the assistant sometimes returns answers that add nothing over a free tool. None of this disqualifies the tool for what it is good at, which is speeding up reading, but it does disqualify it for what people sometimes ask of it, which is closing an analysis on its own.
Pricing and plans
Read the page carefully, because the toggle arrives set to annual, so the first figure you see is not what paying month to month costs.
In that default state the entry plan is Premium, from $12 a month billed annually, with 1,200 credits. If you would rather pay monthly, the same plan is $20. The next step, Advanced, is $70 a month billed annually and $90 monthly, with 10,000 credits. Above that sits Max, sold by credit bracket and starting at $160 a month billed annually, with team rates per user alongside.
There is also a free plan with 100 credits a month, and there is a small naming muddle worth knowing so you do not hunt for it in the wrong place: it does not appear on the pricing page, the official credits guide calls it Basic and the help centre calls it Free. That same guide is what confirms credits expire at the end of each cycle.
A currency warning that saves surprises: the site decides the currency from where you are. From Spain and Latin America it publishes dollars. In India it switches to rupees. And the vendor declares no currency code anywhere, only the glyph, so what you get is what the screen shows from wherever you sit.
It is also worth knowing there is a second, separate price list, the one for the editor inherited from Typeset, with lower figures. It is a different product and the page itself says so, so it is no use for working out what the assistant costs.
Who it is for (and who it is not for)
SciSpace is for people who read scientific papers daily and want to do two things without switching tools: find what has been published and understand what is already in front of them. Doctoral students, people building states of the art, anyone who spends the day with academic PDFs open.
It is not for anyone who wants a predictable budget. If your working pattern is intense and comes in bursts, expiring credits and unpredictable per-task cost play against you, and a flat fee or per-page billing like Humata suits better.
Nor is it for anyone needing an exhaustive, documentable search, or working in the hard sciences and expecting it to settle the analysis. If your job is pulling the same data point from dozens of papers into one comparable table, Elicit is tuned for that. If your material is reports, minutes or contracts rather than papers, look at NotebookLM. And if the documents cannot leave your machine, the answer is AnythingLLM. The full map is in the best AI assistants.
Alternatives to SciSpace
The most useful comparison is against a pure literature-review assistant, which does table extraction better and accompanied reading worse, and against a general notebook, which takes any source and does not bill by credits. The full run-through, with what wins in each case, is in the alternatives to SciSpace.
Frequently Asked Questions
How much does SciSpace cost?
The entry plan, Premium, is $12 a month billed annually and $20 if you pay month to month. There is a free plan with 100 credits a month, which the credits guide calls Basic and the help centre calls Free.
What are credits and why do they matter so much?
They are the unit of consumption for everything it does. They matter because what a task spends depends on the computing power it needs and is unknown until it finishes, and because they expire at the end of each cycle instead of rolling over. There is an indicative table and a real-time ledger, which help but do not predict.
Does it invent references?
It links to the real papers, so checking they exist is immediate. What is worth double-checking is the interpretation, and one engineering professor’s review argues it falls short in the hard sciences.
Can I upload my own PDFs?
Yes, and it can analyse them in batches, which is what separates it from tools built around a single document.
What gets criticised about the billing?
Renewal above all: several verified reviews complain it is charged with no advance reminder unless you cancel. On refunds the accounts do not agree, so the prudent move is reading the terms before taking the annual rate.