Benchmarks › LiveCodeBench

LiveCodeBench

Older benchmarkBy UC Berkeley, MIT and Cornell University ↗Results by Artificial Analysis ↗Since 2024Paper ↗
Solve new programming contest problems
90.8top score · Gemini 3 Flash Preview
152models ranked
111open-weights models
Pass@1higher is better
Oct 20, 2026last updated

How the test works

1

Task

Solve new programming contest problems

  • Kinds of tasksPython contest programming
  • DomainsLeetCode, AtCoder, CodeForces
  • In → outText → Python code
2

Run

Non-reasoning: up to 16,384 output tokens, adjusted to model limits

  • Temperature0 for non-reasoning; 0.6 for reasoning, unless the model lab recommends otherwise
3

Score

Pass@1 (solved on the first try)

  • Graded byCode execution against test cases
  • Better isHigher
Sources: github.com, artificialanalysis.ai

LiveCodeBench · AA archived leaderboardtop 10 of 152

Full leaderboard with filters →
ModelLiveCodeBench · AA archivedPer taskPriceSpeedSupportsContext
Gemini 3 Flash PreviewGoogle
90.8thinking
—$0.50 / $32111M
DeepSeek V3.2 SpecialeDeepSeekOpen
89.6thinking
——128K
GPT-5.2OpenAI
89.4medium
—$1.75 / $14—400K
GLM 4.7Z.aiOpen
89.4thinking
—$0.60 / $2.2061203K
GPT OSS 120BOpenAIOpen
87.8high
$0.092$0.03 / $0.15243131K
Claude Opus 4.5Anthropic
87.1thinking
—$5 / $2558200K
GPT-5.1OpenAI
86.8high
—$1.25 / $10128400K
Kimi K2 ThinkingMoonshot AIOpen
85.3thinking
—$0.47 / $2169256K
GPT-5OpenAI
84.6high
—$1.25 / $10102400K
GPT-5 MiniOpenAI
83.8high
$0.053$0.25 / $2130400K

Versions and boards on Model Menu

BoardModels
LiveCodeBench · AA archivedShown above
153
LiveCodeBench · Official artifact
10

LiveCodeBench questions

What is LiveCodeBench?›

Solve new programming contest problems. Created by UC Berkeley, MIT and Cornell University.

Which model is best on LiveCodeBench?›

Gemini 3 Flash Preview leads LiveCodeBench · AA archived with 90.8, ahead of DeepSeek V3.2 Speciale (89.6) and GPT-5.2 (89.4).

What is the best open-weights model on LiveCodeBench?›

DeepSeek V3.2 Speciale with 89.6, then GLM 4.7 (89.4).

How is LiveCodeBench scored?›

Pass@1 (solved on the first try); code execution against test cases.

Which versions of LiveCodeBench are there?›

LiveCodeBench · AA archived (153 models), LiveCodeBench · Official artifact (10 models).