Skip to content

Benchmark

Agents' Last Exam · Pass Rate

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
16
Scores recorded
23
Sources
1
IQ weight
No weight published

Score distribution

26.7%
Mean
27%
Median
4.4%
Std dev
22.8%
10th percentile
31.8%
90th percentile
11.8%
Min
11.8%35.4%
The band covers the middle 80% of scores. The line marks the median.

Leaderboard

#ModelScore
1DeepSeek V4.1 Flash 1017DeepSeek 1017 first-party evaluation · reasoning effort, candidate sampling, exact run count, and unspecified harnesses not published · published pass rate · harness unspecified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.35.4%
2DeepSeek V4.1 FlashDeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31.8%
3DeepSeek V4.1 Flash官方 scaffold / Pass@1· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31.8%
4GPT-5.6 SolQwen表:Pass Rate· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.30.6%
5Claude Opus 5DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.28.6%
6GLM 5.3ALE-CLI / Pass@1· Reported by Z.ai, DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.28.5%
7GLM 5.3DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.28.5%
8Kimi K3Kimi Code · official leaderboard snapshot 2026-07-23· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.28.3%
9Kimi K3ALE-CLI / Pass@1· Reported by Z.ai, DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.27.6%
10Kimi K3DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.27.6%
11Claude Opus 4.8Qwen表:Pass Rate· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.27%
12Qwen3.8 MaxQwen表:Pass Rate· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.27%
13GPT-5.6 SolDeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.26.7%
14Gemini 3.7 FlashALE-Claw· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.26.3%
15GLM 5.3 Flashofficial ALE · Claude Code max · 1M · 64K output · Tool Search disabled· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.26.3%
16DeepSeek V4 Pro 0813DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.25.7%
17DeepSeek V4 Pro 0813official scaffold · pass@1· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.25.7%
18DeepSeek V4 Flash 0731DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.25.2%
19DeepSeek V4 Flash 0731multimodal inputs explicitly ignored· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.25.2%
20Qwen 3.8 Flash NextQwen3.8-Flash-Next release/model card · Pass@1· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.24.3%
21Hy4 previewCLI· Reported by Tencent Hy· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.22.8%
22Qwen 3.8 27BQwen3.8-27B model card · Pass@1· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.20.4%
23Qwen3.7 MaxQwen表:Pass Rate· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.11.8%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.