What does CyberGym end-to-end (AA) measure?

A full exploit-and-patch cycle in one run, scored end to end. Run by Artificial Analysis. Scores are reported in percent of tasks solved; higher is better.

Which AI model leads CyberGym end-to-end (AA)?

Mistral Large 4 leads CyberGym end-to-end (AA) with 81.7% as of 18 Oct 2026, ahead of MiMo-V2.6-Pro at 78.6%.

How many models have CyberGym end-to-end (AA) results?

18 model configurations have a CyberGym end-to-end (AA) result on BenchLeader, all taken from Artificial Analysis.

Who runs CyberGym end-to-end (AA) and how often is it updated?

CyberGym end-to-end (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.

Does CyberGym end-to-end (AA) count toward the BenchLeader Index?

Yes. CyberGym end-to-end (AA) contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.

CyberGym end-to-end (AA)

A full exploit-and-patch cycle in one run, scored end to end. Run by Artificial Analysis.

As of 18 Oct 2026, Mistral Large 4 leads CyberGym end-to-end (AA) on BenchLeader with 81.7%, ahead of MiMo-V2.6-Pro at 78.6%, across 18 model configurations with a published result.

Published by: Artificial Analysis
Category: Agents & tools
Index weight: 0.5
Models: 18
Data as of: 18 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

Top 15

1Mistral Large 481.7%
2MiMo-V2.6-Pro78.6%
3GPT-6 Luna (max)77.9%
4Step 5 Preview74.8%
5GLM 5.3 Flash74.0%
6Grok 4.7 (xhigh)74.0%
7Kimi K3 (max)58.0%
8Beam 1.045.0%
9GLM 5.3 (max)29.0%
10MiniMax M324.4%
11DeepSeek V4.1 Flash (max)22.9%
12Nemotron 3 Ultra 550B A55B (thinking)10.7%
13Gemini 3.8 Flash (high)8.4%
14Claude Opus 5.5 (max)0.8%
15Claude Fable 5.1 (max)0.8%

18 of 18

#
1	Mistral Large 4Mistral AIopen	81.7%	58.7	2026-10-06
2	MiMo-V2.6-ProXiaomiopen ↗	78.6%	65.3	2026-09-21
3	GPT-6 Luna maxOpenAI	77.9%	58.6	2026-09-22
4	Step 5 PreviewStepFunopen	74.8%	62.2	2026-09-18
5	GLM 5.3 FlashZhipu AIopen ↗	74.0%	62.2	2026-08-26
6	Grok 4.7 xhighSpaceXAI	74.0%	60.6	2026-09-21
7	Kimi K3 maxMoonshot AIopen ↗	58.0%	64.8	2026-07-16
8	NewBeam 1.0Reflectionopen	45.0%	56.4	2026-10-16
9	GLM 5.3 maxZhipu AIopen ↗	29.0%	63.6	2026-08-18
10	MiniMax M3MiniMaxopen ↗	24.4%	56.2	2026-06-01
11	DeepSeek V4.1 Flash maxDeepSeekopen ↗	22.9%	61.5	2026-09-10
12	Nemotron 3 Ultra 550B A55B thinkingNVIDIAopen ↗	10.7%	55.6	2026-06-04
13	Gemini 3.8 Flash highGoogle	8.4%	63.3	2026-09-02
14	Claude Opus 5.5 maxAnthropic	0.8%	71.6	2026-09-22
15	Claude Fable 5.1 maxAnthropic	0.8%	69.9	2026-09-01
16	GPT-6 Astra maxOpenAI	0.0%	69.7	2026-09-03
17	GPT-6 Sol maxOpenAI	0.0%	65.6	2026-09-22
18	Qwen3.8 27B xhighAlibabaopen ↗	0.0%	56.3	2026-08-14

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “CyberGym end-to-end (AA) leaderboard”, https://www.benchleader.com/benchmarks/aa_cybergym_e2e, data as of 18 Oct 2026.

CyberGym end-to-end (AA): questions

What does CyberGym end-to-end (AA) measure?: A full exploit-and-patch cycle in one run, scored end to end. Run by Artificial Analysis. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads CyberGym end-to-end (AA)?: Mistral Large 4 leads CyberGym end-to-end (AA) with 81.7% as of 18 Oct 2026, ahead of MiMo-V2.6-Pro at 78.6%.
How many models have CyberGym end-to-end (AA) results?: 18 model configurations have a CyberGym end-to-end (AA) result on BenchLeader, all taken from Artificial Analysis.
Who runs CyberGym end-to-end (AA) and how often is it updated?: CyberGym end-to-end (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does CyberGym end-to-end (AA) count toward the BenchLeader Index?: Yes. CyberGym end-to-end (AA) contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.