# Expert Question Answering on HLE All Non-Easy

Source: [Expert Question Answering on HLE All Non-Easy](<https://www.sota2.com/research/sota/expert-question-answering-on-hle-all-non-easy>)

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

Data updated: 2026-04-06T07:28:29.216Z

Evaluation of model performance on the 'All Non-Easy' subset of the HLE dataset specifically for Expert Question Answering tasks, reporting metrics such as recovery rate, response length, and compression.

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 | Expert Question Answering |
| Dataset (including any stated split) | HLE All Non-Easy |
| Metric | Recovery Rate |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [Recovery Rate](<https://www.sota2.com/research/sota/expert-question-answering-on-hle-all-non-easy.md?metric=Recovery%20Rate>): Higher is better; unit: PERCENT.

- [Sample Size (N)](<https://www.sota2.com/research/sota/expert-question-answering-on-hle-all-non-easy.md?metric=Sample%20Size%20(N)>): Higher is better; unit: SCALAR.

- [Average Response Length](<https://www.sota2.com/research/sota/expert-question-answering-on-hle-all-non-easy.md?metric=Average%20Response%20Length>): Higher is better; unit: SCALAR.

- [Compression Ratio](<https://www.sota2.com/research/sota/expert-question-answering-on-hle-all-non-easy.md?metric=Compression%20Ratio>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Recovery Rate | Result date | Report paper |
| --- | --- | --- | --- |
| QA+ | 9.3 | Not recorded | 69d33e0c7d0172140ec736b4 |
| QA | 6.5 | Not recorded | 69d33e0c7d0172140ec736b4 |
| Bit-Limited Chain-of-Thought (BL-CoT) | 5.4 | Not recorded | 69d33e0c7d0172140ec736b4 |

### QA+

Recovery Rate: 9.3. Result ID: 69d3609a47f34ce924bacf5f.

[Reporting paper](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736b4.md>)

Recorded evaluation configuration: {"Protocol":"opus→opus→haiku"}. This record does not assert independent reproduction.

### QA

Recovery Rate: 6.5. Result ID: 69d3609a47f34ce924bacf5e.

[Reporting paper](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736b4.md>)

Recorded evaluation configuration: {"Protocol":"haiku→opus→haiku"}. This record does not assert independent reproduction.

### Bit-Limited Chain-of-Thought (BL-CoT)

Recovery Rate: 5.4. Result ID: 69d3609a47f34ce924bacf5d.

[Reporting paper](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736b4.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736b4.md>)

Recorded evaluation configuration: {"Protocol":"haiku→haiku→haiku"}. This record does not assert independent reproduction.

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