benchmark sourcesOfficial benchmark
ALE-Bench
This page's content is in Chinese.
Sakana AI / 编程 · AtCoder 启发式竞赛式的长时算法工程
In AIDC热点Ranked
Evidence budget0.6%
Upstream data timeTo be verified
Last successful sync11/03 08:05
What it measures and how
在没有标准答案的优化题上持续改进程序,按竞赛表现分(Performance)比较。
How to use this evidence
正式计分(方法 v16 起),权重较低;成绩由 Sakana AI 发布、Epoch AI 汇编。
Benchmark results
按固定规则,每个公开模型采用一套代表配置。匿名测试型号不展示,保留原榜名次,每项最多 30 个。
| Original board rank | Original board model | Raw result | Representative config |
|---|---|---|---|
| 1 | gpt-6-astra_maxOpenAI | 2,951.3 | Max 推理 |
| 2 | gpt-6.1-sol_maxOpenAI | 2,743.8 | Max 推理 |
| 3 | gpt-6.1-terra_maxOpenAI | 2,608.8 | Max 推理 |
| 4 | gpt-6-sol_maxOpenAI | 2,462 | Max 推理 |
| 5 | gpt-5.6-sol_maxOpenAI | 2,176.9 | Max 推理 |
| 6 | claude-opus-5_highAnthropic | 2,164.6 | High 推理 |
| 7 | claude-opus-5-5_highAnthropic | 2,147.4 | High 推理 |
| 8 | claude-fable-5-1_highAnthropic | 2,143.2 | High 推理 |
| 9 | gemini-4-argon_highGoogle | 2,100.1 | High 推理 |
| 10 | claude-fable-5_highAnthropic | 2,041.3 | High 推理 |
| 11 | gpt-5.6-terra_maxOpenAI | 1,951.4 | Max 推理 |
| 12 | gpt-5.5_xhighOpenAI | 1,943 | xHigh 推理 |
| 13 | kimi-k3.1-pro_maxMoonshot AI | 1,932.8 | Max 推理 |
| 14 | claude-sonnet-5-5_highAnthropic | 1,819.1 | High 推理 |
| 16 | muse-spark-1.3_xhighMeta | 1,688.4 | xHigh 推理 |
| 17 | gpt-5.6-luna_maxOpenAI | 1,667.4 | Max 推理 |
| 18 | gpt-5.3-codex_xhighOpenAI | 1,655.2 | xHigh 推理 |
| 19 | deepseek-v4.1-flash-1017_maxDeepSeek | 1,622 | Max 推理 |
| 20 | gpt-5.4-2026-03-05_highOpenAI | 1,607 | High 推理 |
| 21 | Muse-Spark-Open-816B-A42B_xhighMeta | 1,590.1 | xHigh 推理 |
| 23 | claude-haiku-5-5_highAnthropic | 1,587 | High 推理 |
| 24 | gpt-6-luna_xhighOpenAI | 1,576.9 | xHigh 推理 |
| 25 | claude-opus-4-8_highAnthropic | 1,563.8 | High 推理 |
| 26 | kimi-k3_maxMoonshot AI | 1,524.5 | Max 推理 |
| 28 | grok-4.6_xhighxAI | 1,508.1 | xHigh 推理 |
| 29 | claude-sonnet-5_highAnthropic | 1,463.1 | High 推理 |
| 31 | deepseek-v4-pro-0813_maxDeepSeek | 1,403.2 | Max 推理 |
| 32 | gemini-3-flash-previewGoogle | 1,367.2 | 来源默认配置 |
| 33 | claude-sonnet-4-6_mediumAnthropic | 1,327.3 | Medium 推理 |
| 34 | claude-opus-4-7Anthropic | 1,323.1 | 来源默认配置 |
Benchmark limits and data attribution
覆盖的前沿模型较少;表现分不是百分比,只用于排序。
Data license:CC BY 4.0 · Epoch AI 汇编,原始成绩由 Sakana AI 发布
成绩由 Sakana AI 发布,原始分数与 AIDC热点 共识分使用不同尺度,不能直接相加。