Runtime Environments

Every submission is compiled and run inside an isolated sandbox on our own judge image. This page lists the exact toolchain behind each language, the flags it is built with, what is already imported for you, and which libraries ship alongside the standard library. The set mirrors LeetCode's environments, so a solution written there runs here unchanged.

Time and memory limits

Every problem declares one time limit and one memory limit, written for C++. Each language multiplies them by a fixed factor so that an interpreted solution is judged against a budget it can meet. The time limit applies to each test case on its own; the memory limit is the peak for the whole run and is capped at 500 MB in every language.

Language	Time factor	Memory factor	On a 2 s / 256 MB problem
Python	5×	2×	10 s / 500 MB
JavaScript	5×	2×	10 s / 500 MB
Java	3×	2×	6 s / 500 MB
C++	1×	1×	2 s / 256 MB
C#	3×	2×	6 s / 500 MB
Go	2×	1×	4 s / 256 MB
Rust	1.5×	1×	3 s / 256 MB
Swift	2×	2×	4 s / 500 MB

Fixed for every language

Stack: 128 MB, so recursion a few hundred thousand frames deep is fine in every language.
Output: 1 MB of stdout per run; more than that is Output Limit Exceeded.
Whole submission: 15 s of CPU time and 20 s of wall-clock time across all test cases, as a backstop behind the per-case limit. A solution that is slow on every case can hit this before any single case times out.
Interview scratch runs (a program with its own stdin, no test cases): 5 s of CPU time and 10 s of wall-clock time.
No network, no filesystem outside the working directory, and no more than a handful of threads or processes.

Python 3.14.7

time 5× · memory 2×

Solutions run under CPython 3.14.7. The whole standard library is available, and the modules interview code reaches for most are imported for you, so a solution can use collections, itertools, heapq, bisect and friends without an import line.

The recursion limit is raised to 550,000 frames and the process stack to 128 MB, so a recursive depth-first search over a 10⁵-node tree does not need an explicit iterative rewrite.

run: python3 script.py

In scope without an import

typing (everything)
collections (module and *)
itertools (module and *)
functools (module and *)
heapq (module and *)
bisect_left, bisect_right, insort, insort_left, insort_right
math (module) plus inf, gcd, lcm, sqrt, isqrt, floor, ceil, log, log2, log10, comb, perm, factorial, prod
string (module) plus ascii_lowercase, ascii_uppercase, ascii_letters, digits
operator (module) plus itemgetter, attrgetter
re, random, copy (module) plus deepcopy
dataclass, field, Enum

Libraries

Package	Version	Provides	Access
sortedcontainers	2.4.0	SortedList, SortedDict, SortedSet	already in scope

Good to know

Solutions run on plain CPython; there is no PyPy option.
No numpy, pandas or other scientific packages — the image ships only sortedcontainers.

Node.js 24.20.0

time 5× · memory 2×

Solutions run under Node.js 24.20.0 LTS, with every modern JavaScript feature the V8 engine of that release supports: optional chaining, nullish coalescing, private class fields, top-level BigInt, Array.prototype.at, structuredClone, and the rest.

The call stack is raised, so deep recursion works the way it does in the other languages instead of throwing "Maximum call stack size exceeded" at a few thousand frames.

run: node script.js (re-executed with --stack-size=100000 before the solution runs)

Libraries

Package	Version	Provides	Access
lodash	4.17.21	_	already in scope
@datastructures-js/priority-queue	6.3.5	PriorityQueue, MinPriorityQueue, MaxPriorityQueue	already in scope
@datastructures-js/heap	4.3.7	Heap, MinHeap, MaxHeap	already in scope
@datastructures-js/queue	4.3.0	Queue	already in scope
@datastructures-js/deque	1.0.8	Deque	already in scope
@datastructures-js/stack	3.1.6	Stack	already in scope
@datastructures-js/linked-list	6.1.4	LinkedList, LinkedListNode, DoublyLinkedList, DoublyLinkedListNode	already in scope
@datastructures-js/set	4.2.2	EnhancedSet	already in scope
@datastructures-js/binary-search-tree	5.4.0	require('@datastructures-js/binary-search-tree')	import it yourself
@datastructures-js/graph	5.3.1	require('@datastructures-js/graph')	import it yourself
@datastructures-js/trie	4.2.3	require('@datastructures-js/trie')	import it yourself

Good to know

The preloaded names are globals, so a solution that declares its own class Queue or const _ simply shadows the library version.
BinarySearchTree, Graph and Trie stay behind an explicit require because problems regularly ask you to implement a class with that name.

Java 25 (OpenJDK 25.0.4.1, Eclipse Temurin)

time 3× · memory 2×

Solutions are compiled with javac 25 and run on the HotSpot JVM from Eclipse Temurin 25.0.4.1 LTS. Records, sealed types, pattern matching for switch, text blocks, var, Stream.toList() and virtual threads are all available.

The commonly used packages are imported for you, and a javafx.util.Pair-compatible Pair<K, V> class (getKey, getValue, equals, hashCode) is defined in scope, as on LeetCode.

compile: javac Main.java
run: java -XX:+UseSerialGC -XX:ActiveProcessorCount=2 -Xms16m -Xmx384m Main

In scope without an import

java.util.*
java.util.concurrent.*
java.util.concurrent.atomic.*
java.util.function.*
java.util.regex.*
java.util.stream.*
java.math.*
java.io.*
Pair<K, V>

Good to know

The heap is capped at 384 MB; an allocation past it is an OutOfMemoryError, not a Memory Limit Exceeded verdict.

C++23 (GCC 14.4.0)

time 1× · memory 1×

Solutions are compiled with g++ 14.4.0 using the GNU dialect of the C++23 standard and libstdc++ from the same release. Code is compiled with level-two optimization (-O2), and <bits/stdc++.h> is included for you through a precompiled header, so every standard header is already in scope.

GNU extensions are available as on most judges: __int128, __builtin_popcount and friends, variable-length arrays, and the policy-based data structures in __gnu_pbds (ordered sets and hash tables).

Ranges, std::span, <format>, <bit>, <numbers>, concepts and the three-way comparison operator all work. Range formatting (std::format("{}", vector)) is a GCC 15 feature and is not available.

compile: g++ -O2 -std=gnu++23 -D_GLIBCXX_ASSERTIONS main.cpp
run: ./a.out

In scope without an import

<bits/stdc++.h>

Good to know

Standard-library assertions are on: an out-of-range operator[] or front() on an empty container aborts with a message naming the operation instead of silently reading bad memory.

C# 14 (.NET 10, SDK 10.0.400)

time 3× · memory 2×

Solutions are compiled by the Roslyn compiler from the .NET 10 SDK (10.0.400) with C# language version 14, optimizations on and nullable reference types disabled, then run on the .NET 10 runtime. The full .NET 10 base class library is available, including PriorityQueue<TElement, TPriority>, System.Numerics.BitOperations, spans, and LINQ.

The most common namespaces are imported for you; anything else needs its own using directive at the top of the solution.

compile: csc -optimize+ -nullable:disable -langversion:latest -warn:0 -out:main.dll main.cs
run: dotnet exec main.dll

In scope without an import

System
System.Collections
System.Collections.Generic
System.Diagnostics
System.Globalization
System.Linq
System.Text

Good to know

The runtime runs in invariant-globalization mode: culture-specific formatting and comparison behave like the invariant culture. Use StringComparer.Ordinal for sorting strings, as you should on any judge.
The heap is capped at 384 MB; an allocation past it is an OutOfMemoryException.

Go 1.27.0

time 2× · memory 1×

Solutions are built with Go 1.27.0, so generics, min, max, clear, range-over-int and the slices, maps and cmp packages all work.

Write import blocks exactly as you would in a normal Go file. The harness lifts your imports into its own file header and merges them with the ones it already needs, so a duplicate import of fmt or sort is never an error.

compile: go build -ldflags="-s -w" main.go
run: ./main

In scope without an import

bufio, bytes, fmt, os
container/heap, container/list
encoding/hex, encoding/json
math, math/big, math/bits, math/rand
regexp, strconv, strings, unicode, time
sort, slices, maps, cmp
runtime/debug

Libraries

Package	Version	Provides	Access
github.com/emirpasic/gods	v1.18.1	github.com/emirpasic/gods/…	import it yourself
github.com/emirpasic/gods/v2	v2.0.0-alpha	github.com/emirpasic/gods/v2/…	import it yourself

Good to know

The gods packages are the only third-party modules available; there is no network access, so nothing else can be imported.

Rust 1.98.0

time 1.5× · memory 1×

Solutions are compiled with rustc 1.98.0 using the 2024 edition with optimizations on. Templates follow the LeetCode shape — an impl Solution block with associated functions — and the harness declares struct Solution; for you unless your code defines it.

The three crates LeetCode ships are available with an ordinary use line; nothing else is on the crate path.

compile: rustc -O --edition 2024 -C strip=symbols main.rs
run: ./prog

In scope without an import

std::rc::Rc
std::cell::RefCell

Libraries

Package	Version	Provides	Access
rand	0.8.8	use rand::…	import it yourself
regex	1.13.1	use regex::…	import it yourself
itertools	0.14.0	use itertools::…	import it yourself

Good to know

Optimizations turn overflow checks off: integer overflow wraps instead of panicking. Reach for i64, checked_* or wrapping_* deliberately rather than relying on either behaviour.

Swift 6.2.4

time 2× · memory 2×

Solutions are compiled with swiftc 6.2.4 with optimizations on. Foundation is imported, and the three Apple packages LeetCode ships are imported for you as well.

Templates use class Solution with underscore argument labels (func twoSum(_ nums: [Int], _ target: Int)), matching LeetCode Swift.

compile: swiftc -O main.swift -o prog
run: ./prog

In scope without an import

Foundation
Collections
Algorithms
Numerics

Libraries

Package	Version	Provides	Access
swift-collections	1.6.0	Deque, Heap, OrderedSet, OrderedDictionary, BitSet, TreeSet, …	already in scope
swift-algorithms	1.2.1	combinations, permutations, chunks, windows, …	already in scope
swift-numerics	1.1.0	Real, Complex, …	already in scope

Good to know

Recursion is as deep as in the other languages; the stack is not the limit.
Int is 64-bit, so problems whose answers exceed 32 bits need no special integer type.
A runtime trap (overflow, out-of-bounds, force-unwrap of nil) ends the whole run, the way a segfault does in C++ — Swift has no per-test-case recovery.

Runtime Environments

Time and memory limits

Language	Time factor	Memory factor	On a 2 s / 256 MB problem
Python	5×	2×	10 s / 500 MB
JavaScript	5×	2×	10 s / 500 MB
Java	3×	2×	6 s / 500 MB
C++	1×	1×	2 s / 256 MB
C#	3×	2×	6 s / 500 MB
Go	2×	1×	4 s / 256 MB
Rust	1.5×	1×	3 s / 256 MB
Swift	2×	2×	4 s / 500 MB

Fixed for every language

Stack: 128 MB, so recursion a few hundred thousand frames deep is fine in every language.
Output: 1 MB of stdout per run; more than that is Output Limit Exceeded.
Whole submission: 15 s of CPU time and 20 s of wall-clock time across all test cases, as a backstop behind the per-case limit. A solution that is slow on every case can hit this before any single case times out.
Interview scratch runs (a program with its own stdin, no test cases): 5 s of CPU time and 10 s of wall-clock time.
No network, no filesystem outside the working directory, and no more than a handful of threads or processes.

Python 3.14.7

time 5× · memory 2×

The recursion limit is raised to 550,000 frames and the process stack to 128 MB, so a recursive depth-first search over a 10⁵-node tree does not need an explicit iterative rewrite.

run: python3 script.py

In scope without an import

typing (everything)
collections (module and *)
itertools (module and *)
functools (module and *)
heapq (module and *)
bisect_left, bisect_right, insort, insort_left, insort_right
math (module) plus inf, gcd, lcm, sqrt, isqrt, floor, ceil, log, log2, log10, comb, perm, factorial, prod
string (module) plus ascii_lowercase, ascii_uppercase, ascii_letters, digits
operator (module) plus itemgetter, attrgetter
re, random, copy (module) plus deepcopy
dataclass, field, Enum

Libraries

Package	Version	Provides	Access
sortedcontainers	2.4.0	SortedList, SortedDict, SortedSet	already in scope

Good to know

Solutions run on plain CPython; there is no PyPy option.
No numpy, pandas or other scientific packages — the image ships only sortedcontainers.

Node.js 24.20.0

time 5× · memory 2×

The call stack is raised, so deep recursion works the way it does in the other languages instead of throwing "Maximum call stack size exceeded" at a few thousand frames.

run: node script.js (re-executed with --stack-size=100000 before the solution runs)

Libraries

Package	Version	Provides	Access
lodash	4.17.21	_	already in scope
@datastructures-js/priority-queue	6.3.5	PriorityQueue, MinPriorityQueue, MaxPriorityQueue	already in scope
@datastructures-js/heap	4.3.7	Heap, MinHeap, MaxHeap	already in scope
@datastructures-js/queue	4.3.0	Queue	already in scope
@datastructures-js/deque	1.0.8	Deque	already in scope
@datastructures-js/stack	3.1.6	Stack	already in scope
@datastructures-js/linked-list	6.1.4	LinkedList, LinkedListNode, DoublyLinkedList, DoublyLinkedListNode	already in scope
@datastructures-js/set	4.2.2	EnhancedSet	already in scope
@datastructures-js/binary-search-tree	5.4.0	require('@datastructures-js/binary-search-tree')	import it yourself
@datastructures-js/graph	5.3.1	require('@datastructures-js/graph')	import it yourself
@datastructures-js/trie	4.2.3	require('@datastructures-js/trie')	import it yourself

Good to know

The preloaded names are globals, so a solution that declares its own class Queue or const _ simply shadows the library version.
BinarySearchTree, Graph and Trie stay behind an explicit require because problems regularly ask you to implement a class with that name.

Java 25 (OpenJDK 25.0.4.1, Eclipse Temurin)

time 3× · memory 2×

The commonly used packages are imported for you, and a javafx.util.Pair-compatible Pair<K, V> class (getKey, getValue, equals, hashCode) is defined in scope, as on LeetCode.

compile: javac Main.java
run: java -XX:+UseSerialGC -XX:ActiveProcessorCount=2 -Xms16m -Xmx384m Main

In scope without an import

java.util.*
java.util.concurrent.*
java.util.concurrent.atomic.*
java.util.function.*
java.util.regex.*
java.util.stream.*
java.math.*
java.io.*
Pair<K, V>

Good to know

The heap is capped at 384 MB; an allocation past it is an OutOfMemoryError, not a Memory Limit Exceeded verdict.

C++23 (GCC 14.4.0)

time 1× · memory 1×

GNU extensions are available as on most judges: __int128, __builtin_popcount and friends, variable-length arrays, and the policy-based data structures in __gnu_pbds (ordered sets and hash tables).

Ranges, std::span, <format>, <bit>, <numbers>, concepts and the three-way comparison operator all work. Range formatting (std::format("{}", vector)) is a GCC 15 feature and is not available.

compile: g++ -O2 -std=gnu++23 -D_GLIBCXX_ASSERTIONS main.cpp
run: ./a.out

In scope without an import

<bits/stdc++.h>

Good to know

Standard-library assertions are on: an out-of-range operator[] or front() on an empty container aborts with a message naming the operation instead of silently reading bad memory.

C# 14 (.NET 10, SDK 10.0.400)

time 3× · memory 2×

The most common namespaces are imported for you; anything else needs its own using directive at the top of the solution.

compile: csc -optimize+ -nullable:disable -langversion:latest -warn:0 -out:main.dll main.cs
run: dotnet exec main.dll

In scope without an import

System
System.Collections
System.Collections.Generic
System.Diagnostics
System.Globalization
System.Linq
System.Text

Good to know

The runtime runs in invariant-globalization mode: culture-specific formatting and comparison behave like the invariant culture. Use StringComparer.Ordinal for sorting strings, as you should on any judge.
The heap is capped at 384 MB; an allocation past it is an OutOfMemoryException.

Go 1.27.0

time 2× · memory 1×

Solutions are built with Go 1.27.0, so generics, min, max, clear, range-over-int and the slices, maps and cmp packages all work.

compile: go build -ldflags="-s -w" main.go
run: ./main

In scope without an import

bufio, bytes, fmt, os
container/heap, container/list
encoding/hex, encoding/json
math, math/big, math/bits, math/rand
regexp, strconv, strings, unicode, time
sort, slices, maps, cmp
runtime/debug

Libraries

Package	Version	Provides	Access
github.com/emirpasic/gods	v1.18.1	github.com/emirpasic/gods/…	import it yourself
github.com/emirpasic/gods/v2	v2.0.0-alpha	github.com/emirpasic/gods/v2/…	import it yourself

Good to know

The gods packages are the only third-party modules available; there is no network access, so nothing else can be imported.

Rust 1.98.0

time 1.5× · memory 1×

The three crates LeetCode ships are available with an ordinary use line; nothing else is on the crate path.

compile: rustc -O --edition 2024 -C strip=symbols main.rs
run: ./prog

In scope without an import

std::rc::Rc
std::cell::RefCell

Libraries

Package	Version	Provides	Access
rand	0.8.8	use rand::…	import it yourself
regex	1.13.1	use regex::…	import it yourself
itertools	0.14.0	use itertools::…	import it yourself

Good to know

Optimizations turn overflow checks off: integer overflow wraps instead of panicking. Reach for i64, checked_* or wrapping_* deliberately rather than relying on either behaviour.

Swift 6.2.4

time 2× · memory 2×

Solutions are compiled with swiftc 6.2.4 with optimizations on. Foundation is imported, and the three Apple packages LeetCode ships are imported for you as well.

Templates use class Solution with underscore argument labels (func twoSum(_ nums: [Int], _ target: Int)), matching LeetCode Swift.

compile: swiftc -O main.swift -o prog
run: ./prog

In scope without an import

Foundation
Collections
Algorithms
Numerics

Libraries

Package	Version	Provides	Access
swift-collections	1.6.0	Deque, Heap, OrderedSet, OrderedDictionary, BitSet, TreeSet, …	already in scope
swift-algorithms	1.2.1	combinations, permutations, chunks, windows, …	already in scope
swift-numerics	1.1.0	Real, Complex, …	already in scope

Good to know

Recursion is as deep as in the other languages; the stack is not the limit.
Int is 64-bit, so problems whose answers exceed 32 bits need no special integer type.
A runtime trap (overflow, out-of-bounds, force-unwrap of nil) ends the whole run, the way a segfault does in C++ — Swift has no per-test-case recovery.