Benchmark leaderboard - 25 entries
LiveCodeBench leaderboard
LiveCodeBench is a holistic and contamination-free evaluation benchmark of LLMs for code that continuously collects new problems over time. As of October 17, 2026, O4-Mini (High) ranks first on the LiveCodeBench board with 80.2.
Last verified from livecodebench.github.io. Read the caveats
LiveCodeBench
25 entries, in the board's order
| 1 | O4-Mini (High) Code generation, Pass@1, rank 1; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 89.4 / Hard 63.5 Official leaderboard | 80.2 |
| 2 | O3 (High) Code generation, Pass@1, rank 2; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 84.4 / Hard 57.1 Official leaderboard | 75.8 |
| 3 | O4-Mini (Medium) Code generation, Pass@1, rank 3; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.2 / Medium 86.5 / Hard 52.7 Official leaderboard | 74.2 |
| 4 | Gemini-2.5-Pro-06-05 Code generation, Pass@1, rank 4; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 87.2 / Hard 50.2 Official leaderboard | 73.6 |
| 5 | DeepSeek-R1-0528 Code generation, Pass@1, rank 5; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.7 / Medium 85.2 / Hard 50.7 Official leaderboard | 73.1 |
| 6 | Gemini-2.5-Pro-05-06 Code generation, Pass@1, rank 6; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.2 / Medium 82.3 / Hard 50.2 Official leaderboard | 71.8 |
| 7 | EXAONE-4.0-32B Code generation, Pass@1, rank 7; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.4 / Medium 82.3 / Hard 46.2 Official leaderboard | 70 |
| 8 | OpenReasoning-Nemotron-32B Code generation, Pass@1, rank 8; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.3 / Medium 81.4 / Hard 46.3 Official leaderboard | 69.8 |
| 9 | O3-Mini-2025-01-31 (High) Code generation, Pass@1, rank 9; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 84.4 / Hard 38.4 Official leaderboard | 67.4 |
| 10 | OpenCodeReasoning-Nemotron-1.1-32B Code generation, Pass@1, rank 10; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 97.9 / Medium 79.6 / Hard 41.1 Official leaderboard | 66.8 |
| 11 | Grok-3-Mini (High) Code generation, Pass@1, rank 11; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 97.3 / Medium 74.5 / Hard 44.8 Official leaderboard | 66.7 |
| 12 | O4-Mini (Low) Code generation, Pass@1, rank 12; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.2 / Medium 80.1 / Hard 38.4 Official leaderboard | 65.9 |
| 13 | Qwen3-235B-A22B Code generation, Pass@1, rank 13; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 80.1 / Hard 37.9 Official leaderboard | 65.9 |
| 14 | XBai-o4-medium Code generation, Pass@1, rank 14; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.2 / Medium 82.3 / Hard 35 Official leaderboard | 65 |
| 15 | O3-Mini-2025-01-31 (Med) Code generation, Pass@1, rank 15; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 80.1 / Hard 31.5 Official leaderboard | 63 |
| 16 | Gemini-2.5-Flash-05-20 Code generation, Pass@1, rank 16; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 71.6 / Hard 35 Official leaderboard | 61.9 |
| 17 | Gemini-2.5-Flash-04-17 Code generation, Pass@1, rank 17; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 70.2 / Hard 33 Official leaderboard | 60.6 |
| 18 | O3-Mini-2025-01-31 (Low) Code generation, Pass@1, rank 18; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 99.1 / Medium 68.8 / Hard 26.1 Official leaderboard | 57 |
| 19 | Claude-Opus-4 (Thinking) Code generation, Pass@1, rank 19; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 98.2 / Medium 70.9 / Hard 24.1 Official leaderboard | 56.6 |
| 20 | Claude-Sonnet-4 (Thinking) Code generation, Pass@1, rank 20; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 97.3 / Medium 66 / Hard 26.6 Official leaderboard | 55.9 |
| 21 | Claude-Sonnet-4 Code generation, Pass@1, rank 21; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 96.4 / Medium 53.9 / Hard 15.8 Official leaderboard | 47.1 |
| 22 | Claude-Opus-4 Code generation, Pass@1, rank 22; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 94.5 / Medium 52.5 / Hard 17.2 Official leaderboard | 46.9 |
| 23 | Claude-3.5-Sonnet-20241022 Code generation, Pass@1, rank 23; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 91.2 / Medium 34.3 / Hard 8.2 Official leaderboard | 36.4 |
| 24 | GPT-4O-2024-08-06 Code generation, Pass@1, rank 24; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 82.5 / Medium 26 / Hard 3.3 Official leaderboard | 29.5 |
| 25 | GPT-4-Turbo-2024-04-09 Code generation, Pass@1, rank 25; default window 8/1/2024 to 5/1/2025 (454 problems, release_v6); Easy 81 / Medium 24.8 / Hard 3.1 Official leaderboard | 28.7 |
About this board
- Staleness warning: the official LiveCodeBench leaderboard (read 2026-10-17) has not been updated since 2025-08-01 (last commit 'add new results' in github.com/LiveCodeBench/livecodebench.github.io). It lists only 28 models and none of the late-2025 or 2026 frontier releases. Entry date is that last-update date, not the read date.
- Scores are for the default time window of 8/1/2024 to 5/1/2025 (454 problems). The leaderboard recomputes when a viewer changes the window, so other windows show different numbers. Ranks 26-28 not recorded: GPT-4O-mini-2024-07-18 27.5, DeepSeek-V3 27.2, Claude-3-Haiku 20.2.
- The description is quoted from the official site (livecodebench.github.io index): 'LiveCodeBench is a holistic and contamination-free evaluation benchmark of LLMs for code that continuously collects new problems over time.'
- Leaderboard link quirks, recorded as displayed: EXAONE-4.0-32B and XBai-o4-medium both link to wenxiaobai.com, and the Claude-Opus-4 rows link to the Claude Sonnet page.
FAQ
What does LiveCodeBench measure?
LiveCodeBench is a holistic and contamination-free evaluation benchmark of LLMs for code that continuously collects new problems over time.
Which model leads LiveCodeBench?
As of October 17, 2026, O4-Mini (High) ranks first on the LiveCodeBench board with 80.2. That is the board's own order, not a Developers Digest ranking.
Where do these LiveCodeBench scores come from?
All 25 entries come from the official leaderboard at livecodebench.github.io. Every entry links to its source, which was read on October 17, 2026, and lists the setting it was run under.
Related reading
- Kolibri Release Guide: Aleph Alpha's 78B Open ModelKolibri is an Apache 2.0 English-German MoE with 78B total and 3B active parameters and a 1M context. Benchmarks, the vLLM command, and who should run it.
- The Oracle Is Agreeing With ItselfSelf-improving test generation loops can grade against an oracle that agrees with itself. An audit shows the inflated gain and why independent checks count.
- Qwen3.8-27B vs Opus 4.6 Max: Local Agentic Coding BenchmarksQwen3.8-27B is a 27B dense Apache-2.0 model that beats Opus 4.6 Max on SWE-bench Pro (61.7 vs 53.4). Benchmarks, hardware, cost and when to run it.
- LFM2.5-2.6B: Liquid AI's On-Device Agent Model Runs at 220 Tokens/s in Under 2.5 GBLFM2.5-2.6B is an open-weight on-device agent model from Liquid AI that runs in under 2.5 GB. How it was trained, what the benchmarks say and how to run it.
- VibeThinker-3B: A 3 Billion Parameter Model That Outscores Opus 4.5 on ReasoningVibeThinker-3B is a 3B parameter reasoning model reporting 94.3% on AIME 2026. What the paper claims, how to run it locally, and what it trades away.
- Fugu Ultra's Frontier Performance Claim, Explained Without the HypeSakana Fugu Ultra claims frontier performance by orchestrating models instead of one giant model. What the benchmarks show, what is novel, what needs proof.
Sources
Last verified . Each row links to the page its score was copied from.
Newsletter
Get new benchmark results as they land
One email per week. New model releases, the leaderboard moves that matter and what they mean for what you build.