# Trajectory-safety diagnosis on ATBench-F

Source: [Trajectory-safety diagnosis on ATBench-F](<https://www.sota2.com/research/sota/trajectory-safety-diagnosis-on-atbench-f>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-13T09:55:52.340Z

Evaluation of trajectory safety diagnosis performance on the ATBench-F dataset, measured using R.S., F.M., and R.H. metrics.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Trajectory-safety diagnosis |
| Dataset (including any stated split) | ATBench-F |
| Metric | R.S. Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 14 of 14 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [R.S. Score](<https://www.sota2.com/research/sota/trajectory-safety-diagnosis-on-atbench-f.md?metric=R.S.%20Score>): Higher is better; unit: SCALAR.

- [F.M. Score](<https://www.sota2.com/research/sota/trajectory-safety-diagnosis-on-atbench-f.md?metric=F.M.%20Score>): Higher is better; unit: SCALAR.

- [R.H. Score](<https://www.sota2.com/research/sota/trajectory-safety-diagnosis-on-atbench-f.md?metric=R.H.%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | R.S. Score | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen2-7B-Instruct + FATE | 49.2 | Not recorded | 6a04059ddbc836b87d8275ac |
| AgentDoG-Qwen2-7B | 46.8 | Not recorded | 6a04059ddbc836b87d8275ac |
| GPT-4 | 33.6 | Not recorded | 6a04059ddbc836b87d8275ac |
| GPT-4o | 29.5 | Not recorded | 6a04059ddbc836b87d8275ac |
| Gemini-1.5-Pro | 24.8 | Not recorded | 6a04059ddbc836b87d8275ac |
| Gemini-1.5-Flash | 18.4 | Not recorded | 6a04059ddbc836b87d8275ac |
| QwQ-32B-Preview | 15.8 | Not recorded | 6a04059ddbc836b87d8275ac |
| Qwen2.5-72B-Instruct | 7.7 | Not recorded | 6a04059ddbc836b87d8275ac |
| Qwen2.5-32B-Instruct | 7 | Not recorded | 6a04059ddbc836b87d8275ac |
| Qwen2.5-7B | 6.6 | Not recorded | 6a04059ddbc836b87d8275ac |
| Llama-3.1-8B-Instruct | 6.2 | Not recorded | 6a04059ddbc836b87d8275ac |
| Qwen2.5-7B-Instruct | 5.3 | Not recorded | 6a04059ddbc836b87d8275ac |
| Qwen2-7B | 4.4 | Not recorded | 6a04059ddbc836b87d8275ac |
| Qwen2-7B-Instruct | 1 | Not recorded | 6a04059ddbc836b87d8275ac |

### Qwen2-7B-Instruct + FATE

R.S. Score: 49.2. Result ID: 6a044a9347f34ce924ea9f21.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

[Implementation](<https://github.com/YinBo0927/FATE>)

Recorded evaluation configuration: {"Model Type":"Ours"}. This record does not assert independent reproduction.

### AgentDoG-Qwen2-7B

R.S. Score: 46.8. Result ID: 6a044a9347f34ce924ea9f20.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/697846dda2b28e761d87c80b.md>)

[Implementation](<https://github.com/AI45Lab/AgentDoG>)

Recorded evaluation configuration: {"Model Type":"Guard models"}. This record does not assert independent reproduction.

### GPT-4

R.S. Score: 33.6. Result ID: 6a044a9347f34ce924ea9f14.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Closed-source models"}. This record does not assert independent reproduction.

### GPT-4o

R.S. Score: 29.5. Result ID: 6a044a9347f34ce924ea9f15.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Closed-source models"}. This record does not assert independent reproduction.

### Gemini-1.5-Pro

R.S. Score: 24.8. Result ID: 6a044a9347f34ce924ea9f17.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Closed-source models"}. This record does not assert independent reproduction.

### Gemini-1.5-Flash

R.S. Score: 18.4. Result ID: 6a044a9347f34ce924ea9f16.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Closed-source models"}. This record does not assert independent reproduction.

### QwQ-32B-Preview

R.S. Score: 15.8. Result ID: 6a044a9347f34ce924ea9f1b.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Qwen2.5-72B-Instruct

R.S. Score: 7.7. Result ID: 6a044a9347f34ce924ea9f18.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Qwen2.5-32B-Instruct

R.S. Score: 7. Result ID: 6a044a9347f34ce924ea9f1c.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Qwen2.5-7B

R.S. Score: 6.6. Result ID: 6a044a9347f34ce924ea9f19.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Llama-3.1-8B-Instruct

R.S. Score: 6.2. Result ID: 6a044a9347f34ce924ea9f1f.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75d11e209131cea6a028.md>)

[Implementation](<https://github.com/meta-llama/llama-agentic-system>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Qwen2.5-7B-Instruct

R.S. Score: 5.3. Result ID: 6a044a9347f34ce924ea9f1e.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Qwen2-7B

R.S. Score: 4.4. Result ID: 6a044a9347f34ce924ea9f1a.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

### Qwen2-7B-Instruct

R.S. Score: 1. Result ID: 6a044a9347f34ce924ea9f1d.

[Reporting paper](<https://www.sota2.com/research/paper/6a04059ddbc836b87d8275ac.md>)

Recorded evaluation configuration: {"Model Type":"Open-source models"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
