Is LeetCode Premium worth it? What 781 locked problems tell you

Search “is LeetCode Premium worth it” or “LeetCode Premium worth it 2026” and one of the results on page one hands you a number that is wrong by about two and a half times. codinginterview.com’s post, dated 2025-10-24 and sitting on page one when I ran the search on 2026-08-20, says Premium gates “300+ exclusive problems, most of which are in the Medium and Hard range”. I pulled LeetCode’s own public problem list the same afternoon, counted the paid_only flag, and got 781.

That gap is why this post exists. codinginterview.com published a round number rather than a count.

What LeetCode Premium actually locks, counted on 2026-08-20

One GET request to leetcode.com/api/problems/all/, then count paid_only grouped by difficulty. Anyone can re-run it in a browser tab. Here is what came back:

So 3,248 problems are free, and nobody finishes 3,248 problems. The headline share matters less than where the paywall sits. Mediums are locked at 21.9% against 13.4% for Easies, and Mediums are what the phone screen and the first onsite round are made of. LeetCode put the gate where the money is.

The editorial claim I had backwards

I started this post expecting to debunk the idea that Premium hides the official solutions, because Two Sum’s editorial opens fine for a logged-out request. One problem is not a sample. So I queried solution{paidOnly canSeeDetail} anonymously across 50 randomly chosen free problems. Thirty-four of them carry an editorial at all, and eleven of those thirty-four come back paidOnly: true. About a third of free problems that have a written solution keep the solution behind the paywall.

Hand-checking the famous ones is worse. Trapping Rain Water, Number of Islands, Minimum Window Substring, Longest Consecutive Sequence, Sudoku Solver, Word Ladder II and LRU Cache all return canSeeDetail: false. Two Sum and Median of Two Sorted Arrays are open. My starting assumption was wrong: the free-problem editorials you actually want are mostly gated, and I would rather correct that here than quietly leave it out.

What did survive the check is the company data. An anonymous query returns null for both companyTags and companyTagStats on every problem I tried, free or paid, with no partial view and no teaser. LeetCode Premium company tags, and the frequency windows attached to them, are the actual product.

So is LeetCode Premium worth it?

Measured on 2026-08-20, Premium locks 781 problems, gates roughly a third of the editorials on free problems, and gates company tags completely, for $35 a month or $159 a year. Buy it if you have a dated interview inside about six weeks at a company you can name. Outside that window you are paying for a decision you have not made yet.

The calendar is doing most of the work in that sentence. In the 2025 Stack Overflow Developer Survey, of 35,451 developers who answered the job-search question, 28.8% were “Considering somewhat” and 14.8% “Considering strongly”, while 8.8% had actually moved voluntarily. Most people reading a post like this sit in the considering bucket. A considering-bucket subscription renews three or four times before it gets used, and the annual plan at $13.25 a month is priced to catch exactly that person.

The annual plan is the wrong default for almost everyone. $159 buys twelve months of a tool whose value concentrates into the three weeks before one specific onsite, and if your search genuinely runs a full year, $159 is not the line item standing between you and an offer. I could be wrong about this. Plenty of people would rather pay once in January and never think about the renewal again, and I would not call that a mistake.

The free alternative decides this

The free GitHub company-tag lists are the substitute people actually reach for, and how much they are worth depends on when they were last updated. I pulled four repos from the GitHub API on 2026-08-20. Stars come from /repos/{owner}/{name}. The commit dates come from /repos/{owner}/{name}/commits, which is the endpoint that matters here, because the repo object’s pushed_at counts branch and tag activity that never touched main.

The hxu296 repo has 11,322 stars and has had nothing land on main since May 2022. Its star count records how useful it was four years ago. Someone comparing Premium against “the free GitHub list” in 2026 and landing there is holding a paid product up against a snapshot of question frequencies from before ChatGPT shipped, and will reasonably conclude that Premium is worth every cent.

Land on liquidslr and the comparison inverts. That repo was updated four days before I checked it, carries 470 company folders, and reproduces Premium’s window structure as five CSVs per company: thirty days, three months, six months, more than six months, all-time, each with a Frequency column. I counted rows in its Google folder: 183 problems in the thirty-day file, 785 in six months, 2,344 all-time. That is structurally the feature you are being asked to pay for, and it is what most people want when they search leetcode company wise questions.

A caveat with teeth. Those repos are scraped from LeetCode’s paid tier and almost certainly breach its terms of service. I am reporting how many stars they have and how stale they get, as evidence about this market. Nothing here is a how-to. The buy decision turns on which link you happen to click.

The LeetCode Premium price, and what I could not confirm

LeetCode’s defaultSubscriptionPricing GraphQL field returned monthlyPrice.originalPrice 35.00 and yearlyPrice.originalPrice 159.00 on 2026-08-20, both with discount 0.0. Annual works out to $13.25 a month, a 62% saving on the monthly rate. A request from an Indian IP returned subscriptionCountry: "in" and the same two figures, so the API surfaced no purchasing-power discount for that region.

Now the caveat, and it covers every number in that paragraph. The API returns bare numerics with no currency field, and every rendered page I tried (subscribe, premium and store) returned HTTP 403 to automated fetches. So the LeetCode Premium price above is first-party but unrendered. USD is an inference, the Indian response says nothing about what a checkout would actually charge, and none of it was confirmed against the page a human sees. Open it yourself before you spend money.

One aside, because it caught me out. LeetCode shipped an official iOS app on 2026-04-10 and updated it to version 1.2.0 on 2026-08-08. The App Store listing advertises searching “the latest company questions”, and Apple’s lookup API reports the app itself as Free, price 0.0. Whether a Premium subscription can be bought inside it, I could not check.

Who should skip it

If your interview is not scheduled, skip it. That covers most people who ask the question and it is the entire answer for them.

Still working through a free curated list? Company tags will make you worse rather than better, because they hand you 2,344 Google-tagged problems when your real problem is that you cannot recognise a sliding window yet. Pattern coverage comes first: the Blind 75, or the NeetCode 150 versus Blind 75 argument if you are still picking between them. Our LeetCode patterns cheat sheet is the free version of that step.

A reason that rarely comes up in the buying decision is that you are interviewing somewhere small, and the mirror data makes it concrete. Company tags need a volume of reported questions before they mean anything. In the liquidslr dataset on 2026-08-20 the all-time file holds 2,344 rows for Google, 2,011 for Amazon and 1,399 for Meta, but only 155 for Zoho and 63 for Airbnb. Razorpay, Postman and Freshworks have no folder at all. Below a certain hiring volume the frequency data you are buying is a handful of rows, and the shape of it will not tell you anything your recruiter would not.

And skip it if your failure mode is performing, not selecting. A randomized controlled trial with 48 Computer Science students, published at ESEC/FSE 2020, found that “performance is reduced by more than half, by simply being watched by an interviewer” (Behroozi, Shirolkar, Barik and Parnin). That study measured whiteboard conditions and I will not stretch it into a claim about subscriptions. But if you can solve Mediums alone and freeze on a call, a better problem list is not your bottleneck, and rehearsing under observation is the thing that moves. That is roughly why we built the LastRound AI copilot around live rounds instead of problem sets. The free tier gives you 15 credits a month, which reset at the start of each month, so you can find out whether the observed-performance gap is your real gap before paying anyone anything.

What I would do

Buy one month, three weeks out from a dated onsite at a company you can name, and cancel the day after. Pull the thirty-day and three-month frequency lists for that company, work them, leave the all-time file alone. The honest version of this question turns out to be a question about your calendar, and a calendar answer does not sell an annual plan.

Every measurement here is cheap to reproduce, which is deliberate. The problem endpoint is undocumented and LeetCode could close it tomorrow, the editorial sample was 50 problems rather than all 3,248, and the repo dates will drift. Re-run the counts before you trust them.