# Turn-Level Crisis Detection on CRADLE-DIALOGUE

Source: [Turn-Level Crisis Detection on CRADLE-DIALOGUE](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue>)

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

Data updated: 2026-06-11T07:35:26.282Z

Evaluation of turn-level crisis detection performance on the CRADLE-DIALOGUE dataset, measured using EM, Jaccard, and F1 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 | Turn-Level Crisis Detection |
| Dataset (including any stated split) | CRADLE-DIALOGUE |
| Metric | EM |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| 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

- [EM](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue.md?metric=EM>): Higher is better; unit: PERCENT.

- [Jaccard Similarity](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue.md?metric=Jaccard%20Similarity>): Higher is better; unit: PERCENT.

- [Micro F1 Score](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue.md?metric=Micro%20F1%20Score>): Higher is better; unit: PERCENT.

- [Macro F1 Score](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue.md?metric=Macro%20F1%20Score>): Higher is better; unit: PERCENT.

- [Micro Recall](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue.md?metric=Micro%20Recall>): Higher is better; unit: PERCENT.

- [Macro Recall](<https://www.sota2.com/research/sota/turn-level-crisis-detection-on-cradle-dialogue.md?metric=Macro%20Recall>): Higher is better; unit: PERCENT.

## Recorded results

| Method | EM | Result date | Report paper |
| --- | --- | --- | --- |
| Claude-4.5-Sonnet | 88.62 | Not recorded | 6a28f31f90966ade919c1622 |
| Gemini-3-Flash | 88.51 | Not recorded | 6a28f31f90966ade919c1622 |
| Qwen3-14B | 87.48 | Not recorded | 6a28f31f90966ade919c1622 |
| gpt-oss-120b | 87.34 | Not recorded | 6a28f31f90966ade919c1622 |
| Qwen3-32b-FT | 86.83 | Not recorded | 6a28f31f90966ade919c1622 |
| Qwen3-32B | 86.33 | Not recorded | 6a28f31f90966ade919c1622 |
| gpt-oss-20b | 86.02 | Not recorded | 6a28f31f90966ade919c1622 |
| GPT-5.1 | 85 | Not recorded | 6a28f31f90966ade919c1622 |
| llama3.3-70b-FT | 84.56 | Not recorded | 6a28f31f90966ade919c1622 |
| Llama-3.3-70B-Instruct | 82.57 | Not recorded | 6a28f31f90966ade919c1622 |
| gemma-3-27b-it | 82.35 | Not recorded | 6a28f31f90966ade919c1622 |
| Qwen2.5-72b-FT | 78.42 | Not recorded | 6a28f31f90966ade919c1622 |
| gemma-3-12b-it | 73.62 | Not recorded | 6a28f31f90966ade919c1622 |
| Llama-4-Scout-17B | 73.34 | Not recorded | 6a28f31f90966ade919c1622 |
| Qwen2.5-72B-Instruct | 70.18 | Not recorded | 6a28f31f90966ade919c1622 |

### Claude-4.5-Sonnet

EM: 88.62. Result ID: 6a2a6532fbc373592c27277e.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

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

### Gemini-3-Flash

EM: 88.51. Result ID: 6a2a6532fbc373592c272780.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

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

### Qwen3-14B

EM: 87.48. Result ID: 6a2a6532fbc373592c272776.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"Qwen Family"}. This record does not assert independent reproduction.

### gpt-oss-120b

EM: 87.34. Result ID: 6a2a6532fbc373592c27277a.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ab01e209131cea84a0c.md>)

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

### Qwen3-32b-FT

EM: 86.83. Result ID: 6a2a6532fbc373592c27277d.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

Recorded evaluation configuration: {"Fine-tuned data":"Dialogue data"}. This record does not assert independent reproduction.

### Qwen3-32B

EM: 86.33. Result ID: 6a2a6532fbc373592c272777.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"Qwen Family"}. This record does not assert independent reproduction.

### gpt-oss-20b

EM: 86.02. Result ID: 6a2a6532fbc373592c272779.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ab01e209131cea84a0c.md>)

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

### GPT-5.1

EM: 85. Result ID: 6a2a6532fbc373592c27277f.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

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

### llama3.3-70b-FT

EM: 84.56. Result ID: 6a2a6532fbc373592c27277b.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

Recorded evaluation configuration: {"Fine-tuned data":"Reddit post-level crisis annotations"}. This record does not assert independent reproduction.

### Llama-3.3-70B-Instruct

EM: 82.57. Result ID: 6a2a6532fbc373592c272772.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

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

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

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"LLaMA Family"}. This record does not assert independent reproduction.

### gemma-3-27b-it

EM: 82.35. Result ID: 6a2a6532fbc373592c272775.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa779f1e209131cea746cd.md>)

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"Gemma Family"}. This record does not assert independent reproduction.

### Qwen2.5-72b-FT

EM: 78.42. Result ID: 6a2a6532fbc373592c27277c.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

Recorded evaluation configuration: {"Fine-tuned data":"Reddit post-level crisis annotations"}. This record does not assert independent reproduction.

### gemma-3-12b-it

EM: 73.62. Result ID: 6a2a6532fbc373592c272774.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa779f1e209131cea746cd.md>)

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"Gemma Family"}. This record does not assert independent reproduction.

### Llama-4-Scout-17B

EM: 73.34. Result ID: 6a2a6532fbc373592c272773.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

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

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

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"LLaMA Family"}. This record does not assert independent reproduction.

### Qwen2.5-72B-Instruct

EM: 70.18. Result ID: 6a2a6532fbc373592c272778.

[Reporting paper](<https://www.sota2.com/research/paper/6a28f31f90966ade919c1622.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa767e1e209131cea6d6a7.md>)

Recorded evaluation configuration: {"Category":"Open-source Model","Family":"Qwen Family"}. This record does not assert independent reproduction.

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