Skip to content

Benchmark

CyberGym Level 1 success rate

Higher scores are better.

Models measured
31
Scores recorded
37
Sources
1
IQ weight
1.53

Score distribution

0.444
Mean
0.432
Median
0.296
Std dev
0.072
10th percentile
0.818
90th percentile
0.0186
Min
0.01860.863
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.53 for chat models, measured over 31 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Gemini 3.8 Flash Cyberagent: Google Antigravity, trials: 1, reported by: Google, date: 2026-09-10· CyberGym leaderboard data docs0.863
2GPT-5.5-cyberagent: OpenAI Agent, trials: 1, reported by: OpenAI, date: 2026-06-22· CyberGym leaderboard data docs0.856
3GLM 5.3agent: Claude Code, trials: 1, reported by: Zhipu AI, date: 2026-08-14· CyberGym leaderboard data docs0.845
4DeepSeek V4 Proagent: DeepSeek Agent, trials: 1, reported by: DeepSeek, date: 2026-08-13· CyberGym leaderboard data docs0.833
5GPT 5.5agent: OpenAI Agent, trials: 1, reported by: OpenAI, date: 2026-04-23· CyberGym leaderboard data docs0.818
6Grok 4.6agent: Grok Build, trials: 1, reported by: xAI, date: 2026-08-12· CyberGym leaderboard data docs0.797
7GPT 5.4agent: OpenAI Agent, trials: 1, reported by: OpenAI, date: 2026-04-23· CyberGym leaderboard data docs0.79
8Grok 4.5agent: Grok Build, trials: 1, reported by: xAI, date: 2026-07-14· CyberGym leaderboard data docs0.79
9GLM 5.2agent: Claude Code, trials: 1, reported by: Zhipu AI, date: 2026-07-21· CyberGym leaderboard data docs0.772
10DeepSeek V4 Flashagent: DeepSeek Agent, trials: 1, reported by: DeepSeek, date: 2026-07-31· CyberGym leaderboard data docs0.767
11GLM 5.1agent: Claude Code, trials: 1, reported by: Zhipu AI, date: 2026-04-07· CyberGym leaderboard data docs0.687
12Claude Opus 4.6agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2026-02-05· CyberGym leaderboard data docs0.666
13GPT 5.4agent: Codex CLI, trials: 1, reported by: Zhipu AI, date: 2026-04-07· CyberGym leaderboard data docs0.663
14Claude Sonnet 4.6agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2026-02-17· CyberGym leaderboard data docs0.652
15Muse Spark 1.1agent: Meta Agent, model note: helpful-only, trials: 1, reported by: Meta AI, date: 2026-07-09· CyberGym leaderboard data docs0.59
16DeepSeek-V4-Pro Previewagent: Claude Code, trials: 1, reported by: XDxAI, date: 2026-07-13· CyberGym leaderboard data docs0.577
17Claude Opus 4.5agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2026-02-05· CyberGym leaderboard data docs0.506
18Muse Sparkagent: Meta Agent, trials: 1, reported by: Meta AI, date: 2026-04-14· CyberGym leaderboard data docs0.435
19GLM 5agent: Claude Code, trials: 1, reported by: Zhipu AI, date: 2026-02-12· CyberGym leaderboard data docs0.432
20Kimi K2.5agent: Kimi Agent, trials: 1, reported by: Kimi, date: 2026-02-02· CyberGym leaderboard data docs0.413
21GPT 5agent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-12-05· CyberGym leaderboard data docs0.394
22Gemini 3.1 Pro Previewagent: Gemini CLI, trials: 1, reported by: Zhipu AI, date: 2026-04-07· CyberGym leaderboard data docs0.388
23Claude Sonnet 4.5agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2025-09-29· CyberGym leaderboard data docs0.289
24Claude Opus 4.1agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2025-09-29· CyberGym leaderboard data docs0.25
25GLM 4.7agent: Claude Code, trials: 1, reported by: Zhipu AI, date: 2026-02-12· CyberGym leaderboard data docs0.235
26Claude Sonnet 4agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2025-09-29· CyberGym leaderboard data docs0.226
27Claude Sonnet 4agent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-23· CyberGym leaderboard data docs0.179
28Claude Sonnet 3.7agent: Anthropic Agent, trials: 1, reported by: Anthropic, date: 2025-09-29· CyberGym leaderboard data docs0.145
29Claude Sonnet 3.7agent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.119
30GPT 4.1agent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0936
31GPT 4.1agent: Cybench, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0896
32GPT 4.1agent: Codex CLI, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0737
33GPT 4.1agent: ENiGMA, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0723
34Gemini 2.5 Flash Previewagent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0484
35DeepSeek V3agent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0358
36o4 Miniagent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0246
37Qwen 3 235B A22Bagent: OpenHands, trials: 1, reported by: CyberGym Team, date: 2025-05-15· CyberGym leaderboard data docs0.0186

Data credits: CyberGym leaderboard (Wang, Shi, He, Cai, Zhang, Song; UC Berkeley), ICLR 2026. Results evaluated and submitted by the listed teams. No licence is stated for the result data; the repository's Apache-2.0 licence covers its code only.