Meta’s interview process, as US new grads describe it in 2025, is a recruiter screen, a CodeSignal assessment for some, a 45-minute technical screen, then three onsite interviews of two LeetCode-style problems each plus a behavioral round on a prompt such as resolving a conflict. Code isn’t run, so candidates find their own bugs and state complexity. Experienced loops add an AI-enabled coding round and system design.
Several 2025 new grads with no offer said the questions were hard for the level or that a hiring freeze ended their process, so it pays to keep other applications moving while you wait.
Meta interview rounds, as candidates report them
| Round | Duration | What it tests |
|---|---|---|
| Recruiter screen | 20-30 min | Background, motivation, logistics |
| Online Assessment (CodeSignal, some candidates) | 70-90 min | Progressive multi-stage coding problem, video/mic monitored |
| Technical phone screen | 45 min | 1 live coding problem |
| Onsite - Coding (traditional) | 45 min | DSA - arrays/strings, graphs, trees, intervals |
| Onsite - Coding (AI-enabled, since late 2025) | ~60 min | Coding with a built-in AI assistant; your judgment and prompting, not the assistant’s output |
| Onsite - System design / architecture | 45-60 min | Scaled to level - lighter architecture chat for E3, full distributed design for E4+ |
| Onsite - Behavioral | 45 min | 5 signal areas: driving results, ambiguity, communication, growth, conflict resolution |
| Team matching | Varies | Team/org placement after the offer |
Meta interview process for E3 new grads: 2025 reports
These are short candidate reports from Taro, not full write-ups, so read them as patterns.
- Menlo Park, May 2025, accepted offer. Three interviews plus one follow-up round. All three were LeetCode-style, with two medium-to-hard problems each. The one behavioral question shared was how they resolved a conflict.
- US (Menlo Park, Seattle and New York), June 2025, no offer. An online assessment, then one behavioral round and two technical rounds. The behavioral prompt was to describe a time they solved a hard problem. They never heard back and put it down to a hiring freeze.
- Menlo Park, October 2025, no offer. LeetCode-style questions solved without running the code; they had to find their own bugs and state time and space complexity. They said Meta’s top-tagged LeetCode questions were a good guide.
- New York, September 2025, no offer. The candidate did not list rounds but called it a lot of rounds with hard questions for a new grad role.
Meta program manager and AI coding reports, 2025-2026
- Program manager, August 2025 (no offer). A 30-minute HR screen, a 45-minute hiring manager interview on Zoom, then five 45-minute Zoom interviews. Every interviewer asked “tell me about a time” questions and “why Meta?” Two months in total.
- Program manager, San Francisco, 2026 (no offer). A hiring manager round, then four 45-minute interviews, each on a different program management area, including “Tell me about a time you influenced without direct authority.” About six weeks.
- Program manager, 2025 (offer accepted). Mostly cross-functional coordination: past projects in detail, what went wrong, and how the candidate adapted communication for different groups. One prompt was a project managed end to end.
- AI-enabled coding, from October 2025. Hello Interview’s write-up of candidate reports describes one 60-minute problem in three phases (fix a bug, implement, optimize) and says the AI felt less helpful in the real interview than in the practice environment.
Meta interview questions that come up often
Coding / technical
- Merge Intervals - sorting + interval merging
- Longest Substring Without Repeating Characters - sliding window + hashmap
- Binary Tree Vertical Order Traversal - BFS with column tracking
- Number of Islands - grid BFS/DFS
- Group Anagrams - hashmap of sorted keys
- Product of Array Except Self
- Minimum Remove to Make Valid Parentheses
- Diameter of a Binary Tree
System design
- Design Facebook/Instagram News Feed (ranking, caching, real-time updates)
- Design Messenger or a chat/notification system at scale
- Design a search or content-ranking service
- Design a rate limiter or distributed cache
- Design a food-delivery or local-search service (geospatial indexing, Elasticsearch-style retrieval)
Meta AI coding interview (AI-enabled round, 2025+)
- Debug and extend a multi-file codebase (maze/graph traversal, serialization)
- Maze solver that prints the path; maximize unique characters from a word list; three cards summing to 15; friend recommendations (problems reported by Hello Interview)
- Implement BFS/DFS extensions with tests provided in-environment
- Explain and verify any AI-suggested change before accepting it
Behavioral
- Highest-impact project you’ve shipped - how did you measure it?
- A decision made with incomplete information, or something that broke after moving fast
- Direct/difficult feedback given or received
- A time you failed and what you changed
- A disagreement with a manager or teammate and how it resolved
What Meta candidates wish they had known
- The CodeSignal OA is a real gate for many candidates now (added in 2025) - practice under a monitored, timed setting, not just untimed LeetCode.
- If you draw the AI-enabled coding round, write a first pass yourself before leaning on the assistant, and be ready to explain why you accepted or rejected its suggestions.
- Scale your system-design answer to your target level - E3 loops often get a lighter architecture chat, not a full distributed-systems design.
- Don’t treat the behavioral round as a formality because it feels conversational - public reports say it (along with system design) decides your final level, sometimes independent of how well coding went.
- Prepare a distinct story for each of Meta’s five signals (driving results, ambiguity, communication, growth, conflict resolution) rather than reusing one story for all of them.
Meta’s behavioral round covers what an HR conversation would elsewhere: motivation, career direction and stories against the five signals. STAR sample answers for each signal are on the Meta behavioral questions page.
Sample answers for high-frequency questions
Q. How do you solve Merge Intervals?
A. Sort the intervals by start time, then sweep once keeping a current interval. For each next interval, if its start is at most the current interval’s end they overlap, so extend the current end to the maximum of the two ends; otherwise push the current interval to the result and make the next one current. Push the final interval after the loop. Sorting dominates at O(n log n) time with O(n) output space. The follow-ups Meta interviewers reach for are inserting a single new interval into an already-sorted list, which is O(n) with a three-phase scan, and handling touching intervals such as [1,3] and [3,5], where you must ask whether they should merge.
Q. Minimum Remove to Make Valid Parentheses - what is the approach?
A. Do two passes with a stack of indices. In the first pass, scan left to right: push the index of each open bracket; on a close bracket, pop if the stack is non-empty, otherwise record that close bracket’s index as unmatched. After the pass, whatever indices remain on the stack are unmatched open brackets. Collect both sets, then build the result string skipping those indices. That is O(n) time and O(n) space, and it removes the minimum number of characters because every index marked was provably unmatchable. A common mistake is deleting from the string during the scan, which shifts indices - mark first, rebuild once.
Q. How do you do a vertical order traversal of a binary tree?
A. Run a BFS while carrying a column index: the root is column 0, a left child is column minus one, a right child is column plus one. Append each node’s value into a hash map keyed by column, and track the minimum and maximum column seen. Because BFS visits level by level, values land in top-to-bottom order automatically, which is why BFS is preferred over DFS here - a DFS needs an explicit sort by row afterwards. At the end, emit the columns from minimum to maximum. Time is O(n) and space O(n). The clarifying question worth asking is how to order two nodes in the same row and column: the harder variant requires sorting those values, the classic Meta version accepts left-to-right insertion order.
Q. How would you count the number of islands in a grid?
A. Scan every cell; when you find an unvisited land cell, increment the island counter and flood-fill from it with DFS or BFS, marking every reachable land cell visited so it is never counted again. Each cell is visited a constant number of times, so it is O(rows times columns) time; space is O(rows times columns) in the worst case for the recursion stack or queue. Mutating the input grid in place (writing a 0 over visited land) avoids a separate visited matrix, but you should say out loud that you are modifying the caller’s data and offer to restore it. Typical follow-ups are switching to 8-directional connectivity, handling a grid too large for memory by streaming rows, and the union-find variant used when islands are added one at a time.
Q. Product of Array Except Self, without division.
A. Build the answer from two directional prefix products. First pass left to right: result[i] holds the product of everything before i, maintained by a running variable starting at 1. Second pass right to left: keep a running suffix product, multiply it into result[i], then update it with the original nums[i]. That gives O(n) time and O(1) extra space beyond the output array, which is what the interviewer is looking for. Division is banned partly because a single zero anywhere breaks it; the two-pass approach handles zeros naturally, since one zero leaves exactly one non-zero entry and two zeros make everything zero.
Frequently asked questions
Do Meta new grads get system design?
Not usually. 2025 US new grad reports on Taro describe three LeetCode-style interviews with two medium-to-hard problems each, plus a behavioral round. System design or product architecture appears from E4.
Can you run your code in a Meta coding interview?
Not in the technical screen or the classic coding rounds. A Menlo Park new grad in October 2025 had to find their own bugs and state time and space complexity without running the code. The AI-enabled coding round is different: it runs in a CoderPad project with tests you can execute.
What did a Meta program manager loop look like in 2025-2026?
Glassdoor reports describe a recruiter call, a 45-minute hiring manager interview, then four or five 45-minute video interviews, each on one area of program management. Every interviewer asked tell me about a time questions, and one asked about influencing without direct authority.
Meta India: a Bengaluru interview experience
Entry-level hiring at Meta’s Bengaluru office runs through University Grad / early-career postings, a summer internship track, and referrals or off-campus applications - not a college-wide placement cell. No fixed CGPA cutoff is published, and there’s no evidence of a TCS/Infosys-style multi-college drive. Treat a forwarded “Meta is coming to our campus” message as unverified until you check Meta’s careers page directly.
Software Engineer E4, Bengaluru (2025, no offer)
Source: InterviewExperiences.in, November 2025
The candidate, an SDE-2 at another large tech company, was approached by a Meta recruiter on LinkedIn.
Screening
- Online assessment: one problem that grew over four or five stages, built around a database cache using maps and an LRU structure. They cleared every stage except the last.
- 45-minute coding screen with a Meta engineer: two problems, both solved, though the first was unfamiliar and took a long time. The recruiter said they had only marginally passed, because the OA was not perfect and the coding was rated “OK”.
Onsite (four rounds, two per day, about two weeks after the screen)
- Coding (45 min): two problems in the style of the screen. Problems the candidate listed from the loop include Diagonal Traverse II, Next Permutation, Sum Root to Leaf Numbers and a sliding-window median.
- AI-assisted coding (60 min): a multi-part problem like the OA, with a weak AI assistant. The task was to pick three cards summing to 15 from 16 cards, replacing them each time, over 12 turns. They fixed a failing unit test but did not finish the main problem.
- Behavioral (45 min): conflict, influence, a project they were proud of, a difficult coworker, a change of direction, a weakness and a production bug.
- Product design (45 min): a food delivery website, with follow-ups on geospatial indexing, Elasticsearch for text search and scheduled jobs for delivery notifications.
Result: Rejected three days after the onsite. The recruiter said only that one round had a negative rating; the candidate believed it was the AI-assisted coding round.
Sources
- Taro: Meta new grad, Menlo Park, May 2025
- Taro: Meta new grad, US, June 2025
- Taro: Meta new grad, Menlo Park, October 2025
- Taro: Meta new grad, New York, September 2025
- Simplify: Meta new grad guide
- Glassdoor: Meta Program Manager interview reviews
- Hello Interview: Meta’s AI-enabled coding interview
- InterviewExperiences.in: Meta E4 Bangalore
The Meta interview process overview covers PM, data engineer and data scientist loops, pay and the full FAQ.