HierSVA-B - Proof Success: leaderboard
HierSVA-B (2026) benchmarks LLM-generated SystemVerilog assertions for hierarchical hardware formal verification on 342 BaseJump STL modules, checked by formal proof with a commercial tool. This board is the share of generated assertions, in percent, that are proven non-vacuously; higher is better.
Metric: Assertion proof success rate (%, C2-P): share of generated assertions in evaluable runs that are proven non-vacuously on HierSVA-DS (342 hierarchical BaseJump STL modules, depths 0 to 9): static generation of SystemVerilog Assertions from RTL and submodule contracts, checked by formal proof in VC Formal; one run, reasoning enabled, temperature 0, strict JSON output. Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Max (Preview) (Thinking) | 95.2 | |
| 2 | GLM-5.1 | 91 | #179 |
| 3 | Gemini 3.1 Pro (Preview) | 90.3 | #94 |
| 4 | Claude Opus 4.7 (Thinking) | 87.5 | #79 (Claude Opus 4.7) |
| 5 | Kimi K2.6 | 86.2 | #178 |
| 6 | GPT-5 Mini | 77.9 | #348 |
| 7 | Qwen 3.6 Plus (Thinking) | 76.9 | #222 (Qwen 3.6 Plus) |
| 8 | GPT-5.5 | 76.8 | #49 |
| 9 | Claude Haiku 4.5 (Thinking) | 74 | #568 (Claude Haiku 4.5) |
| 10 | MiniMax-M2.7 | 58.2 | #518 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hiersva-b-proof-success · How It Works · Data refreshed daily, snapshot 2026-11-13.