ALE-Bench
Long-horizon algorithm engineering on AtCoder heuristic contests.
As of 19 Sept 2026, GPT-5.6 Sol leads ALE-Bench on BenchLeader with 2176.9, ahead of Claude Opus 5 at 2164.6, across 110 model configurations with a published result.
- Published by
- Epoch AI Benchmarking Hub
- Category
- Coding
- Index weight
- Reference only
- Models
- 110
- Data as of
- 19 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
The model iterates on a heuristic-contest solution over hours, scored like a contestant.
How it is scored
Contest performance rating, as published by ALE-Bench.
What to keep in mind
Rating scale like competitive programming; not a percent.
- 1GPT-5.6 Sol (max)2176.9
- 2Claude Opus 5 (high)2164.6
- 3Claude Fable 5.1 (high)2143.2
- 4Claude Fable 5 (high)2041.3
- 5GPT-5.6 Terra (max)1951.4
- 6GPT-5.5 (xhigh)1943.0
- 7GPT-5.6 Luna (max)1667.4
- 8GPT-5.3 Codex (xhigh)1655.2
- 9GPT-5.4 (high)1607
- 10GPT-5.5 (medium)1589.4
- 11Claude Opus 4.8 (high)1563.8
- 12Kimi K3 (max)1524.5
- 13GPT-5.4 (medium)1520.7
- 14Grok 4.6 (xhigh)1508.1
- 15Claude Sonnet 5 (high)1463.1
110 of 110
| # | ||||
|---|---|---|---|---|
| 1 |  GPT-5.6 SolmaxOpenAI | 2176.9 | 68.8 | 2026-07-09 |
| 2 |  Claude Opus 5highAnthropic | 2164.6 | 70.2 | 2026-07-24 |
| 3 | Claude Fable 5.1highAnthropic | 2143.2 | 72.0 | 2026-09-01 |
| 4 | Claude Fable 5highAnthropic | 2041.3 | 64.0 | 2026-06-09 |
| 5 | GPT-5.6 TerramaxOpenAI | 1951.4 | 65.0 | 2026-07-09 |
| 6 | GPT-5.5xhighOpenAI | 1943.0 | 67.6 | 2026-04-23 |
| 7 | GPT-5.6 LunamaxOpenAI | 1667.4 | 60.2 | 2026-07-09 |
| 8 | GPT-5.3 CodexxhighOpenAI | 1655.2 | 65.4 | 2026-02-05 |
| 9 | GPT-5.4highOpenAI | 1607 | 59.0 | 2026-03-05 |
| 10 | GPT-5.5mediumOpenAI | 1589.4 | 64.6 | 2026-04-23 |
| 11 | Claude Opus 4.8highAnthropic | 1563.8 | 62.2 | 2026-05-28 |
| 12 |  Kimi K3maxMoonshot AIopen ↗ | 1524.5 | 67.2 | 2026-07-16 |
| 13 | GPT-5.4mediumOpenAI | 1520.7 | 55.9 | 2026-03-05 |
| 14 |  Grok 4.6xhighxAI | 1508.1 | 64.6 | 2026-08-12 |
| 15 | Claude Sonnet 5highAnthropic | 1463.1 | 62.2 | 2026-06-30 |
| 16 | Claude Opus 4.8no reasoningAnthropic | 1411.8 | – | 2026-05-28 |
| 17 |  DeepSeek V4 Pro 0813maxDeepSeekopen ↗ | 1403.2 | 56.0 | 2026-08-13 |
| 18 |  Gemini 3 FlashGoogle | 1367.2 | 57.7 | 2025-12-17 |
| 19 | Claude Sonnet 4.6mediumAnthropic | 1327.3 | 52.0 | 2026-02-17 |
| 20 | Claude Opus 4.7Anthropic | 1323.0 | 64.5 | 2026-04-16 |
| 21 |  GLM-5.3highZhipu AIopen ↗ | 1317.4 | – | 2026-08-14 |
| 22 | Grok 4.5highxAI | 1309.3 | 61.0 | 2026-07-08 |
| 23 | DeepSeek V4 Flash 0731maxDeepSeekopen ↗ | 1306.1 | 54.4 | 2026-07-31 |
| 24 | GPT-5.2-CodexOpenAI | 1299.9 | 54.0 | 2025-12-18 |
| 25 | GPT-5.2highOpenAI | 1293.5 | 55.7 | 2025-12-11 |
| 26 | GPT-5.2mediumOpenAI | 1249.8 | 58.7 | 2025-12-11 |
| 27 | GPT-5.1-CodexOpenAI | 1244.9 | – | 2025-11-12 |
| 28 | GPT-5.1-Codex-MaxmaxOpenAI | 1208.8 | – | 2025-11-19 |
| 29 | GPT-5.1highOpenAI | 1192.2 | 58.7 | 2025-11-13 |
| 30 |  Qwen3 7maxAlibaba | 1189.4 | 61.0 | 2026-05-19 |
| 31 | GPT-5.4 minihighOpenAI | 1188.6 | 53.0 | 2026-03-17 |
| 32 | Gemini 3 ProGoogle | 1176.8 | 61.1 | 2025-11-18 |
| 33 | GPT-5highOpenAI | 1162.5 | 58.6 | 2025-08-07 |
| 34 | Gemini 3.1 ProGoogle | 1160.6 | 63.9 | 2026-02-19 |
| 35 | Grok 4.20xAI | 1150.3 | – | 2026-02-17 |
| 36 | GPT-5.5no reasoningOpenAI | 1127.6 | 54.0 | 2026-04-23 |
| 37 | Kimi K2.6Moonshot AIopen ↗ | 1092.7 | 60.5 | 2026-04-20 |
| 38 | GPT-5.4no reasoningOpenAI | 1086.0 | 53.4 | 2026-03-05 |
| 39 | GLM-5.2highZhipu AIopen ↗ | 1047.0 | – | 2026-06-16 |
| 40 | Claude Opus 4.5Anthropic | 1025.4 | 58.2 | 2025-11-24 |
| 41 | GLM-5.2maxZhipu AIopen ↗ | 1010.2 | 63.9 | 2026-06-16 |
| 42 | Deepseek v4 ProhighDeepSeek | 1006.1 | 60.8 | 2026-04-24 |
| 43 | GPT-5.4 nanohighOpenAI | 1004.5 | 49.1 | 2026-03-17 |
| 44 | Claude Opus 4.6Anthropic | 996.5 | 63.7 | 2026-02-05 |
| 45 | Grok 4.3xAI | 944.2 | 50.5 | 2026-04-17 |
| 46 | o3highOpenAI | 933.5 | 53.0 | 2025-04-16 |
| 47 | Gemma 4 26B A4BGoogle | 927.2 | 57.1 | 2026-04-02 |
| 48 | Gemma 4 31BGoogleopen ↗ | 925.5 | 52.7 | 2026-04-02 |
| 49 | Gemini 3.5 FlashhighGoogle | 911.0 | 63.6 | 2026-05-19 |
| 50 | Gemini 3.7 FlashhighGoogle | 904.3 | 64.6 | 2026-08-13 |
| 51 |  MiMo-V2.5-ProXiaomiopen ↗ | 899.8 | 59.2 | – |
| 52 | GLM-5.1Zhipu AIopen ↗ | 887.1 | 56.9 | 2026-04-07 |
| 53 | Kimi K2.7 CodeMoonshot AIopen ↗ | 886.2 | 55.9 | 2026-06-12 |
| 54 | o4-minihighOpenAI | 826.2 | 50.8 | 2025-04-16 |
| 55 | Kimi K2.5Moonshot AIopen | 821.6 | 53.9 | 2026-01-27 |
| 56 | GPT-5minimalOpenAI | 807.6 | 46.0 | 2025-08-07 |
| 57 | DeepSeek R1 0528DeepSeekopen | 804.1 | 50.2 | 2025-05-28 |
| 58 | GPT-5 minihighOpenAI | 799.8 | 55.3 | 2025-08-07 |
| 59 | Gemini 3.1 Flash LiteGoogle | 797.7 | 54.5 | 2026-03-03 |
| 60 | Claude Sonnet 4.5Anthropic | 796.1 | 54.3 | 2025-09-29 |
Cite as: BenchLeader, “ALE-Bench leaderboard”, https://www.benchleader.com/benchmarks/ale_bench, data as of 19 Sept 2026.
ALE-Bench: questions
- What does ALE-Bench measure?
- The model iterates on a heuristic-contest solution over hours, scored like a contestant. Scores are reported in score; higher is better.
- Which AI model leads ALE-Bench?
- GPT-5.6 Sol leads ALE-Bench with 2176.9 as of 19 Sept 2026, ahead of Claude Opus 5 at 2164.6.
- How many models have ALE-Bench results?
- 110 model configurations have a ALE-Bench result on BenchLeader, all taken from Epoch AI Benchmarking Hub.
- Who runs ALE-Bench and how often is it updated?
- ALE-Bench is published by Epoch AI Benchmarking Hub. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does ALE-Bench count toward the BenchLeader Index?
- No. ALE-Bench is shown for reference but left out of the composite index.