Here's the detail most Blind 75 guides skip. The engineer who built the list, Yangshun Tay, a former Meta staff engineer, stopped recommending it in May 2022. He published a successor called Grind 75 and called it, in his own words, a better version of Blind 75. So when a guide tells you the Blind 75 is the definitive list, keep in mind the person who made it moved on three years ago.

That doesn't make the Blind 75 useless. It's still one of the most efficient ways to cover the patterns that show up in real coding rounds. But the framing matters, and it changes who should reach for the original list versus the newer one. Let's get into the actual mechanics.

What the Blind 75 actually is

The Blind 75 is a curated set of 75 LeetCode problems, grouped by topic, that Tay posted on the Blind app (the anonymous workplace network) around 2018. The pitch was simple. Instead of grinding 500 random problems, solve these 75 and you'll have seen most of the patterns that companies like Google, Amazon, Meta, and Stripe reuse. The list skews medium-difficulty. A handful of easy problems, a wall of mediums, and a small set of hards that teach the trickier patterns.

The 75 break down across roughly ten topic buckets. Here's the shape of it.

Notice trees and DP swallow a third of the list between them. That's not an accident. Those two topics generate the largest number of interview variations, so the payoff per problem is higher there than in, say, the binary section.

Why the creator replaced it

When Tay built Grind 75, he didn't just reshuffle the old set. He pulled the top LeetCode questions by popularity and frequency, ranked 169 of them, then picked the strongest 75 with a difficulty-ordered progression instead of a topic-grouped one. The reason matters more than the new problem list.

The original Blind 75 is grouped by topic. You do all the trees, then all the graphs. That feels tidy. It's also a worse way to learn, because real interview prep needs interleaving. When you solve fourteen tree problems back to back, your brain starts pattern-matching to "this is a tree problem" before you've even read it. In an actual round, nobody tells you the category. So Grind 75 orders problems by difficulty and mixes topics, which is closer to how recall works under pressure. That's a real improvement, and it's the single best argument for not using the original list as-is.

A study order that survives contact with a real schedule

The old "do it in 6 to 8 weeks at 2 to 3 hours a day" plan assumes you have 2 to 3 free hours a day. Most people don't. Here's a version weighted toward learning value per hour, not tidy topic blocks.

One rule that matters more than the order. After you solve a problem, wait three days and solve it again from a blank file. If you can't, you memorized it, you didn't learn it. That gap is the whole game.

Where the Blind 75 quietly falls short

I think the Blind 75 is overrated for senior interviews, and underrated for the part of the round it doesn't touch at all. Two specific gaps.

First, it's a 2018-era list. It's light on patterns that have gotten more common since, like monotonic stacks, more involved backtracking, and the harder graph variants that show up at the staff level. If you're interviewing for L5 and up, the Blind 75 alone undersamples the top of the difficulty range. It's a floor, not a ceiling.

Second, and this is the bigger one, the list trains exactly one skill: arriving at a correct solution. Coding rounds don't grade only that. They grade whether you can talk through your approach, respond to a curveball follow-up, and recover when the interviewer pushes on your time complexity. The 2024 Stack Overflow Developer Survey, which gathered 65,536 responses, found that the top way developers learn is online resources and documentation, not live, spoken practice. That maps to what goes wrong in interviews. People drill silently against a list, then freeze the first time they have to say their plan out loud to another human.

What we hear from candidates using the copilot

Candidates who run the LastRound AI copilot during live coding rounds tend to fall into a pattern we see again and again. The ones who grind a list like the Blind 75 in silence get the code right and then go quiet. Long pauses. No narration. The interviewer can't tell if they're stuck or thinking, and the silence reads as the former.

The shift that seems to help most isn't more problems. It's talking while solving, the way you would in the room. Candidates tell us the awkward part was never the algorithm, it was saying "I'm going to try a hash map here, and if the input's sorted I'll swap to two pointers" out loud without feeling like they're performing. The Blind 75 builds the first half of that sentence. It does nothing for the second. That's the half that decides close rounds.

So who is it actually for?

If you're early in prep, interviewing for new-grad through mid roles, and you want a finite list you can finish, the Blind 75 (or Grind 75, which I'd pick first) is a good use of your time. The finiteness is the point. A defined list you can complete beats an infinite queue you'll abandon.

If you're senior, treat it as a warm-up, not the plan. Layer in harder patterns and spend real time on the spoken part of the round. And if you've already done most of the Blind 75, I wouldn't restart on Grind 75 just because it's newer. Finish what you started and put the saved hours into mock rounds where you talk. That's the honest version of the advice. I don't have clean data on how this plays out for non-engineering roles like PMs or data scientists, so I'd take the senior-IC framing above as engineer-specific.