Coding Interview Difficulty Is 60% Medium. Your Prep Isn't.
- ~60% of coding interview problems at Amazon, Meta, and Google are medium difficulty, making mediums the core of any serious prep plan.
- Medium problems generate more interviewer signal than easy or hard: they have a brute-force layer, an optimized layer, and edge cases to surface in 45 minutes.
- Phone screens pull from the easier half of the medium band; later onsite rounds push toward the harder half, sometimes touching low-hard territory.
- Senior and staff interviews raise the bar, not the question: the same medium demands tradeoff discussion, complexity precision, and alternative proposals.
- Hards earn returns only after 50+ mediums in a category; before that, they mostly teach you that hards are hard.
- Solve mediums cold with a 35-minute timer, no tags, no hints, then debrief explicitly on what structural signal you missed.
You're grinding LeetCode hards at midnight, telling yourself this is what separates candidates. Meanwhile, Amazon's actual problem pool is 60% medium. You're training for a marathon and the race is a 5K.
The distribution from LeetCode-tagged problem sets for the three most-studied companies:
The Data
| Company | Easy | Medium | Hard |
|---|---|---|---|
| Amazon | 19% | 60% | 21% |
| Meta | 26% | 60% | 14% |
| 27% | 52% | 21% |
Meta, often talked about as one of the most demanding engineering bars, asks hard problems 14% of the time. Amazon, which has a reputation for brutal grind, is one-in-five hards. Not one-in-two.
Google is the real outlier: highest hard proportion, most balanced distribution. If Google is your target, you need a harder edge in your prep. For everyone else:
Across all three, the ~60% medium number is stable. The medium tier isn't the plurality. It is the interview.
Why Mediums Beat Hards for Practice
A hard problem under time pressure is binary. You've seen the insight before, or you haven't. And if you haven't, you're spending 35 minutes staring at the screen doing your best impression of a competent person while the interviewer quietly writes "no hire" in their notes.
Mediums have a brute-force layer, an optimized layer, and edge cases in between, giving the interviewer four or five checkpoints in 45 minutes.
That's why mediums exist at this frequency. More signal per minute than either extreme. It also explains why the 45-minute format locked in everywhere. A medium problem fits cleanly inside that window. The format and the difficulty tier co-evolved. This is not a coincidence.
If you're seeing the hard problem's insight for the first time in your interview, you're not solving it. You're auditioning for a documentary about software engineering anxiety.
Phone Screen vs Onsite: Not the Same Medium
The aggregate distribution smooths over something important. Phone screens pull from the easier end of the medium band. Onsites push toward the harder end.
Think of medium not as one difficulty but a spectrum: "medium-easy" (recognize the pattern, apply it) to "medium-hard" (combine two patterns, extend the approach, explain the tradeoff while writing code and making eye contact). Phone screens live in the lower half. Later onsite rounds live in the upper half, sometimes dipping into low-hard territory.
Prepare for mediums across the full spectrum, not just the approachable end. The ones that took you 45 minutes cold, where your first approach was completely wrong, are the onsite reps.
Senior and Staff: Same Questions, Different Scorecard
Coding difficulty doesn't shift as dramatically as people expect at senior levels. You're not walking into a room of hards. What shifts is the bar, not the question.
At mid-level, solving the medium correctly is enough. At senior, the interviewer expects you to recognize the solution, explain the tradeoff, discuss complexity with precision, propose an alternative, and handle follow-up constraints. Same question. Different scorecard. The medium just became a final exam.
The real difficulty increase at senior comes from system design rounds displacing coding rounds, not from coding rounds getting harder.
How to Actually Allocate Prep Time
If interviews are ~60% medium and you're working through 50 problems, that's roughly 30 mediums, 9 hards, and 12 easies. Most people's actual distribution looks nothing like that. It looks like 12 easies (felt productive), 8 mediums (got stuck, felt bad), and 30 hards (gave up after 10 minutes on each, felt like a fraud).
- Easies are vocabulary, not practice. Build fluency in the primitives, then stop.
- Hards earn returns after 50+ mediums in a category. Before that, a hard mostly teaches you that hards are hard.
- Solve mediums cold with a timer. No tags, no hints, 35 minutes, then debrief on what structural signal you missed. That's the rep that transfers.
Weight by target: more hards for Google, fewer for Meta, full medium range for Amazon since 21% hard is real.
The Takeaway
The difficulty distribution is the syllabus. Companies publish it in every tagged problem set, and it's been stable for years. Mediums are the exam. Easies are the warm-up. Hards are the bonus round for people who have already done the exam.
Prep like the data says to, not like the anxiety says to.
If you want to pressure-test whether your medium problem-solving holds under real conditions, SpaceComplexity runs voice-based mock interviews where you have to explain your thinking out loud, not just produce a solution.
Related: Easy Problems Don't Build Interview Intuition. LeetCode Mediums Do. | You're Practicing LeetCode Wrong, and It's Costing You