Benchmarks · Agents & computer use
HoneyBench
Score = 1 - the published overall hack rate on scored runs. Refusals are not counted as hacks. Score: 1 - overall hack rate. Collected by Jake Boggs' benchmark dashboard.
22 models · 22 results · data as of 2026-11-15 · goodhartlabs.com ↗
Leaderboard
Fable 5.2
All results
22 rows| # | Model | Effort | HoneyBench |
|---|---|---|---|
| 1 | Fable 5.2 | 88.1% | |
| 2 | Opus 5.5 | 68.4% | |
| 3 | Sonnet 5.5 | 58.4% | |
| 4 | GPT-6 Astra | 49.2% | |
| 5 | Inkling 2 Small | 47.4% | |
| 6 | GPT-6.1 Sol | 40.5% | |
| 7 | Fable 5.1 | 38.2% | |
| 8 | Gemini 4 Argon | 36.4% | |
| 9 | Step 5 Preview | 35.1% | |
| 10 | Qwen 3.8 Max | 33.8% | |
| 11 | DeepSeek V4 Pro | 28.6% | |
| 12 | Kimi K3.1 | 28.2% | |
| 13 | DeepSeek V4.1 Flash 1017 | 23.6% | |
| 14 | GLM 5.3 | 23.2% | |
| 15 | Kimi K3 | 21.5% | |
| 16 | Gemini 3.8 Flash | 19.5% | |
| 17 | Muse Wonder | 19.0% | |
| 18 | DeepSeek V4.1 Flash | 19.0% | |
| 19 | Mistral Large 4 | 16.4% | |
| 20 | Muse Spark 1.3 | 16.4% | |
| 21 | Grok 4.7 | 14.9% | |
| 22 | Muse Spark Open | 14.4% |