# Which company interviews are hardest, and what they pay

Source: https://www.techinterview.org/post/3233476093/which-company-interviews-are-hardest-and-what-they-pay/
Updated: 2026-07-02 · techinterview.org

Netflix usually decides on a senior engineer in about four conversations. Amazon will run you through six or seven. The round count tells you almost nothing about which is harder to pass, and even less about what the offer looks like once it lands.

Two things move on their own here: how hard it is to get a yes, and what that yes is worth. A company can be brutal to get through and still pay in the middle of the market. Another can pass more people and out-pay everyone, because it hands you cash instead of a four-year equity cliff. Track those separately and interview "difficulty" stops being a single hard-to-easy slider.

## The loops, side by side

This is the rough shape of the loop at the companies people ask about most, plus the part that actually sinks candidates and the general posture on pay. Read the pay column as direction, not a quote. Real numbers by level come further down.

| Company | Onsite rounds | Where people actually fail | Pay posture |
| --- | --- | --- | --- |
| Google | 4–5, then a hiring committee | Optimal solution plus a clear running commentary; the committee packet, not any one interviewer | Top of market at senior, equity-heavy |
| Meta | 4–5, fast turnaround | Speed (two mediums in ~40 minutes) and getting down-leveled to E4 | Top of market, strong refreshers |
| Amazon | 5–7, including a Bar Raiser | Behavioral depth against the Leadership Principles, not the coding | Base capped early, RSUs backloaded to years 3–4 |
| Apple | 5–8, run by the team | Unpredictability; each org sets its own bar | Strong, varies a lot by org |
| Netflix | ~4, senior only | The judgment bar; there is no junior rung to land on | All cash, top of market, you split salary vs options |
| Microsoft | 4–5, "as appropriate" | A calmer bar; thin system-design answers at senior level | Just under the FAANG top at the same level, closing the gap |
| Stripe | 4–5, including an integration or bug round | Production realism; code that has to compile and run | Strong; weigh the private-company equity |
| OpenAI / Anthropic | Take-home or long pairing, plus values | Mission fit and research-adjacent judgment | Very high; equity in units with their own liquidity rules |
| Jane Street / Citadel Securities | 4–6, math-heavy | Probability, mental math, market-making games | Highest new-grad numbers anywhere |

## Round count is a bad proxy for difficulty

Recruiters and ex-interviewers who have shared numbers put the big-tech offer rate at somewhere near 1 to 3 percent of applicants, and roughly 15 to 25 percent of the people who reach the onsite. That second figure is the one worth staring at. By the time you are in the loop, three out of four still walk away without an offer, and the reason is rarely a missed edge case.

Netflix runs about four rounds and is one of the hardest yeses in the industry, because there is no junior title to fall back to. Get down-leveled at Google and you still have an offer; there is no soft landing at a company that only hires senior and above. Amazon runs almost twice as many rounds, and a large share of that time is behavioral, mapped line by line to the Leadership Principles. Plenty of people with clean coding rounds fail Amazon on "tell me about a time" answers that stay shallow.

Down-leveling deserves its own mention, because it quietly changes the math. Meta and Google will both extend an offer a level below what you targeted rather than reject you outright. That counts as a pass in the stats and a disappointment in your inbox, sometimes a $60k-to-$100k swing in total comp for the same year of work. The interview was not "easy" because you got an offer. You may have gotten the offer for the job below the one you wanted.

The quant shops sit at the far end. Public pass rates barely exist, but the new-grad funnels at Jane Street, Citadel Securities, and Hudson River Trading are widely described as tighter than any FAANG loop, and the content is different in kind: expected value, conditional probability, and live market-making games rather than tree traversals.

## What the offer actually looks like

Numbers move with the stock price, so treat anything specific as a snapshot and check [levels.fyi](https://www.levels.fyi) for the current read on the exact company and level. As of mid-2026, the shape is roughly this. A mid-level engineer (Google L4 and its equivalents) commonly lands in the $240k to $350k total-comp range, with a median near $300k. Senior (L5) more often sits in the $340k to $450k band, and staff and above runs well past that. Base salary is a smaller part of the story than people expect, often $180k to $210k, with the rest in equity and bonus.

Structure is where companies diverge, and it matters more than the headline. Amazon caps cash compensation and backloads restricted stock, vesting 5 percent, then 15 percent, then 40 and 40 across four years, with sign-on bonuses papering over the thin first two years. Take the Amazon number at face value and year one looks great; year three is where the equity actually shows up, and only if the stock held. Netflix goes the other way and pays close to the whole package in cash, letting you decide how much salary to convert into stock options. That removes the vesting-cliff risk and removes the equity upside in the same move.

The AI labs pay at or above the top of that range, but the equity is in profit units or similar instruments rather than public shares, so the liquidity question is real and worth asking directly. Quant is the outlier for new grads: all-in first-year packages at the top market-makers are regularly described in the $300k to $450k+ area, which is staff-level big-tech money handed to someone a year out of school. That is the trade for a pass rate that makes FAANG look forgiving.

## How to read a specific loop before you show up

Ask the recruiter for the round breakdown in writing, the target level, and whether a system-design round is included. Those three answers tell you what to practice and roughly where the bar sits. A loop with two coding rounds and no design is a mid-level loop; a design round means they are testing you at senior, and a weak design answer will down-level you even when the coding is clean.

Then match the prep to the house style instead of grinding problems in general. A few representative prompts, phrased close to how they land in the room:

- Meta: "Merge k sorted lists. Now the lists live on different machines, so walk me through how you'd shard it."

- Amazon: "Tell me about a time you disagreed with your manager and turned out to be right." (scored against Have Backbone and Disagree and Commit)

- Jane Street: "I flip a fair coin until I see two heads in a row. What is the expected number of flips?"

- Stripe: "Here is a failing webhook handler and a flaky test. Make the suite green."

Each of those rewards a different kind of preparation. Drilling LeetCode mediums does nothing for the Amazon prompt, and rehearsing STAR stories does nothing for the coin problem. Look up the specific team on Blind and levels.fyi before you build a plan, because the same company name covers very different loops depending on the org and the level.

The version of this that gets people hired is unglamorous. Find out which rounds you are walking into and what level they map to. Learn how the pay is structured if you pass. Then prepare to that specific loop instead of to a reputation, because a company's difficulty and a company's offer are two different questions, and the people who negotiate well are the ones who answered both before the first call.