HAL Reliability - Overall: leaderboard
Metric: Overall reliability (0-1, mean of consistency, predictability and robustness; higher is better). Source: hal.cs.princeton.edu. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 0.87 |
| 2 | Claude Opus 4.7 | 0.86 |
| 3 | Gemini 3.5 Flash | 0.85 |
| 4 | Gemini 3.1 Pro (Preview) | 0.84 |
| 5 | Claude Sonnet 4 | 0.84 |
| 6 | GPT-5.5 | 0.84 |
| 7 | O1 | 0.8 |
| 8 | Gemini 2.5 Pro | 0.8 |
| 9 | GPT-5.2 (Medium) | 0.77 |
| 10 | Claude 3.5 Haiku | 0.76 |
| 11 | GPT-5.2 | 0.76 |
| 12 | GPT-4 Turbo | 0.74 |
| 13 | Gemini 2.5 Flash | 0.73 |
| 14 | GPT-4o Mini | 0.72 |
| 15 | Claude 3 Haiku | 0.69 |
Interactive version: theaggregate.ai/benchmark?slug=hal-reliability-overall · How It Works · Data refreshed daily, snapshot 2026-09-19.