TS-Guard Release Strict - ASB-Traj - F1: leaderboard
TS-Bench step-level label classification. Strict/loose/exact mappings are distinct tasks; accuracy andF1 measure correctness; recall alone lacks falsepositive penalty and is retained as sourcecontext. These are not modelrefusal rates. AgentHarm/ASB/AgentDojo trajectorypopulations remain separate. Metric: F1 score (%). Higher is better. Source: https://arxiv.org/html/2601.10156v1
Metric: F1 score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 63.03 | #339 |
| 2 | Qwen 2.5 7B Instruct | 62.96 | #857 |
| 3 | Qwen 3 8B | 38.93 | #671 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=ts-guard-release-strict-asb-traj-f1 · How It Works · Data refreshed daily, snapshot 2026-10-17.