SecGoal (Few-Shot): leaderboard
SecGoal (2026) is an expert-annotated benchmark for extracting formalizable security-goal statements from the specification documents of 15 widely deployed protocols. This board is Extraction-Coverage F1 from 0 to 1, the harmonic mean of extraction precision and property-level recall, with few-shot prompting, averaged over protocols; higher is better.
Metric: Extraction-Coverage F1 (0-1), the harmonic mean of extraction precision and property-level recall, for extracting formalizable security-goal statements from protocol documents with few-shot prompting, averaged over 15 protocols; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 0.496 | #123 |
| 2 | GLM-5 | 0.496 | #276 |
| 3 | Qwen 3.5 Plus | 0.49 | #250 |
| 4 | Gemini 3 Flash (Preview) | 0.432 | #156 |
| 5 | DeepSeek V4 Pro | 0.404 | #201 |
| 6 | DeepSeek V4 Flash | 0.4 | #273 |
| 7 | GPT-5.4 | 0.368 | #138 |
| 8 | Claude Sonnet 4.6 | 0.336 | #179 |
| 9 | Qwen 2.5 7B Instruct | 0.326 | #2732 |
| 10 | Gemma 2 9B (IT) | 0.226 | #2383 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=secgoal-few-shot · How It Works · Data refreshed daily, snapshot 2026-11-22.