SkillTrustBench v1.1 - AIG - Suspicious F1: leaderboard
SkillTrustBench v1.1 Suspicious F1 for nine identity-bound model runs inside Tencent AI-Infra-Guard. The fixed public subset has 556 agent-skill cases: 316 malicious, 74 suspicious and 166 normal. Native project verdicts are preferred; empty reports count normal and missing verdicts may use severity fallback. The scored content manifest is pinned; exact framework revision, reasoning effort and budget are unreported. The ambiguous DeepSeek 0910/generic-route row is held separately.
Metric: Suspicious F1 (0–1). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.71 | |
| 2 | DeepSeek V4.1 Flash (1017) (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.68 | |
| 3 | Claude Haiku 5.5 (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.65 | |
| 4 | GLM-5.3 (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.64 | |
| 5 | MiMo-V2.6-Flash (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.55 | |
| 6 | Hy4 preview (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.53 | |
| 7 | MiniMax-M3 (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.52 | |
| 8 | GPT-6 Luna (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.46 | |
| 9 | MiniMax-M3.1-Flash (SkillTrustBench Run: AIG v1.1; 2026-10-25; effort and budget unreported) | 0.43 |
Interactive version: theaggregate.ai/benchmark?slug=skilltrustbench-v1-1-aig-suspicious-f1 · How It Works · Data refreshed daily, snapshot 2026-11-05.