PLE Hidden Test - Overall - hermes: leaderboard
PLE evaluates agents on 56 hidden tasks from professional engineering work. Results distinguish the agent tools and engineering discipline. Mean task reward (0–1).
Metric: Mean task reward (0–1). Source: ple-bench.com. Saturation forecast: Around July 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 0.47 | #3 |
| 2 | Claude Sonnet 5.5 | 0.45 | #12 |
| 3 | Claude Opus 5 | 0.4 | #15 |
| 4 | Gemini 4 Argon | 0.39 | #9 |
| 5 | Claude Fable 5 | 0.33 | #19 |
| 6 | Grok 4.7 | 0.26 | #27 |
| 7 | GLM-5.2 | 0.24 | #77 |
| 8 | Gemini 3.8 Flash | 0.17 | #25 |
| 9 | Gemini 3.1 Pro (Preview) | 0.15 | #61 |
| 10 | Qwen 3 Coder 480B A35B | 0.06 | #279 |
| 11 | MiniMax-M2 | 0.06 | #322 |
| 12 | DeepSeek V4.1 Flash | 0.05 | #38 |
| 13 | Kimi K2 (Thinking) | 0.05 | #246 (Kimi K2) |
| 14 | GPT-OSS-120B | 0.04 | #341 |
| 15 | Qwen 3 Next 80B A3B Instruct | 0.02 | #375 |
No result here: #5 GPT-6 Astra, #7 GPT-6.1 Sol, #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=ple-hidden-test-overall-hermes · How It Works · Data refreshed daily, snapshot 2026-10-29.