Frontier AI Risk Monitor - MASK: leaderboard
The model is pressured by a system prompt to state something it otherwise affirms is false; the number is the platform's MASK honesty score (share of 4,438 items without a lie). This is Concordia AI's own re-run on 2025-2026 models, distinct from Scale's MASK leaderboard and the MASK paper tables.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 78 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 97.68 | #6 |
| 2 | Claude Opus 4.5 (Thinking) | 96.39 | #156 (Claude Opus 4.5) |
| 3 | Claude Sonnet 4 (Thinking) | 95.47 | #400 (Claude Sonnet 4) |
| 4 | Claude Sonnet 4.5 (Thinking) | 95.39 | #250 (Claude Sonnet 4.5) |
| 5 | GPT-5.6 Sol | 94.03 | #30 |
| 6 | Grok 4.7 | 92.72 | #44 |
| 7 | Claude Opus 5.5 | 92.52 | #5 |
| 8 | GPT-5.4 (High) | 91.87 | #123 (GPT-5.4) |
| 9 | Claude Opus 4.8 | 90.98 | #56 |
| 10 | MiniMax-M2.7 | 88.28 | #518 |
| 11 | GPT-5.5 | 87.98 | #49 |
| 12 | GPT-5.1 (High) | 87.65 | #229 (GPT-5.1) |
| 13 | GLM-5 | 86.95 | #260 |
| 14 | Claude Opus 4.6 (Thinking) | 84.74 | #102 (Claude Opus 4.6) |
| 15 | MiniMax-M3 | 81.55 | #269 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-mask · How It Works · Data refreshed daily, snapshot 2026-11-13.