# Steerability on MFQ-30

Source: [Steerability on MFQ-30](<https://www.sota2.com/research/sota/steerability-on-mfq-30>)

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

Data updated: 2026-04-21T10:45:46.622Z

Evaluation of model steerability performance on the MFQ-30 dataset, measuring metrics related to bias and fairness (Harm, Fair, Ingr, Auth, Pur, Avg).

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 | Steerability |
| Dataset (including any stated split) | MFQ-30 |
| Metric | Harm Score |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 6 of 6 |
| 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

- [Harm Score](<https://www.sota2.com/research/sota/steerability-on-mfq-30.md?metric=Harm%20Score>): Lower is better; unit: SCALAR.

- [Fairness Score](<https://www.sota2.com/research/sota/steerability-on-mfq-30.md?metric=Fairness%20Score>): Higher is better; unit: SCALAR.

- [Ingr Score](<https://www.sota2.com/research/sota/steerability-on-mfq-30.md?metric=Ingr%20Score>): Higher is better; unit: SCALAR.

- [Auth Score](<https://www.sota2.com/research/sota/steerability-on-mfq-30.md?metric=Auth%20Score>): Higher is better; unit: SCALAR.

- [Pur Score](<https://www.sota2.com/research/sota/steerability-on-mfq-30.md?metric=Pur%20Score>): Higher is better; unit: SCALAR.

- [Average Steerability Score](<https://www.sota2.com/research/sota/steerability-on-mfq-30.md?metric=Average%20Steerability%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Harm Score | Result date | Report paper |
| --- | --- | --- | --- |
| Claude 3 Haiku | 0.138 | Not recorded | 68fa77dc1e209131cea759fe |
| Claude 3 Sonnet | 0.216 | Not recorded | 68fa77dc1e209131cea759fe |
| LLaMA 3 70B | 0.22 | Not recorded | 68fa77dc1e209131cea759fe |
| Claude 3 Opus | 0.292 | Not recorded | 68fa77dc1e209131cea759fe |
| GPT-3.5 Turbo | 0.296 | Not recorded | 68fa77dc1e209131cea759fe |
| GPT-4o | 0.463 | Not recorded | 68fa77dc1e209131cea759fe |

### Claude 3 Haiku

Harm Score: 0.138. Result ID: 69e753a747f34ce924ce52ad.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77dc1e209131cea759fe.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Claude 3 Sonnet

Harm Score: 0.216. Result ID: 69e753a747f34ce924ce52ac.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77dc1e209131cea759fe.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### LLaMA 3 70B

Harm Score: 0.22. Result ID: 69e753a747f34ce924ce52ab.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77dc1e209131cea759fe.md>)

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

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

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Claude 3 Opus

Harm Score: 0.292. Result ID: 69e753a747f34ce924ce52b0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77dc1e209131cea759fe.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GPT-3.5 Turbo

Harm Score: 0.296. Result ID: 69e753a747f34ce924ce52ae.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77dc1e209131cea759fe.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727a1e209131cea56a1e.md>)

[Implementation](<https://github.com/openai/evals>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GPT-4o

Harm Score: 0.463. Result ID: 69e753a747f34ce924ce52af.

[Reporting paper](<https://www.sota2.com/research/paper/68fa77dc1e209131cea759fe.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

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