BoundaryBench (NIST High): leaderboard
BoundaryBench success on all 89 Terminal-Bench 2.1 tasks, three trials per model-harness bundle at high reasoning effort, under the NIST-derived high least-privilege policy. Success is percent of passing trials, higher is better; policy severity and model reasoning effort are separate settings. The paper reports the Claude Fable 5 gated route with its declared Opus 4.8 fallback: across all 801 trials over three policies, 27.6% of trials touched the fallback and 29.9% of requests used it; these pooled shares are not policy-specific.
Metric: Success Rate under NIST high policy (%). Source: boundarybench.com. Saturation forecast: Around June 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.5 (High) | 74.9 | #54 (Grok 4.5) |
| 2 | GPT-5.6 Sol (High) | 74.2 | #22 (GPT-5.6 Sol) |
| 3 | Claude Fable 5 (High) | 67.8 | #20 (Claude Fable 5) |
| 4 | Gemini 3.7 Flash (High) | 66.7 | #24 (Gemini 3.7 Flash) |
| 5 | Claude Opus 5 (High) | 65.2 | #16 (Claude Opus 5) |
| 6 | Claude Opus 4.8 (High) | 63.3 | #39 (Claude Opus 4.8) |
| 7 | GPT-5.6 Terra (High) | 63.3 | #40 (GPT-5.6 Terra) |
| 8 | GLM-5.2 (High) | 55.8 | #80 (GLM-5.2) |
| 9 | Claude Sonnet 5 (High) | 55.1 | #65 (Claude Sonnet 5) |
| 10 | Kimi K3 (High) | 54.7 | #35 (Kimi K3) |
| 11 | GPT-5.6 Luna (High) | 53.9 | #79 (GPT-5.6 Luna) |
| 12 | MiniMax-M3 (High) | 44.2 | #151 (MiniMax-M3) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=boundarybench-nist-high · How It Works · Data refreshed daily, snapshot 2026-11-05.