# Scientific Question Answering on HLE-Verified Gold (test) Source: [Scientific Question Answering on HLE-Verified Gold (test)]() Collection index: [llms.txt]() Data updated: 2026-06-02T11:27:04.711Z Evaluation of model performance on the test split of the HLE-Verified Gold dataset for Scientific Question Answering tasks, measured by Accuracy. 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 | Scientific Question Answering | | Dataset (including any stated split) | HLE-Verified Gold (test) | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 12 of 12 | | 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 - [Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | ATLAS-MM | 60 | Not recorded | 6a1e63e0dbd7150506f854fd | | Self-Refine (no early stop) | 58 | Not recorded | 6a1e63e0dbd7150506f854fd | | ATLAS | 56 | Not recorded | 6a1e63e0dbd7150506f854fd | | Self-Refine | 53 | Not recorded | 6a1e63e0dbd7150506f854fd | | GPT-5.2-H | 52.48 | Not recorded | 6a1e63e0dbd7150506f854fd | | Reward-model reranking | 52 | Not recorded | 6a1e63e0dbd7150506f854fd | | Budget Forcing | 51 | Not recorded | 6a1e63e0dbd7150506f854fd | | Opus-4.6 | 50.16 | Not recorded | 6a1e63e0dbd7150506f854fd | | Gemini-3-Pro | 48.93 | Not recorded | 6a1e63e0dbd7150506f854fd | | Qwen3-Max-T | 48.48 | Not recorded | 6a1e63e0dbd7150506f854fd | | Opus-4.5 | 48.16 | Not recorded | 6a1e63e0dbd7150506f854fd | | Pass@1 | 48 | Not recorded | 6a1e63e0dbd7150506f854fd | ### ATLAS-MM Accuracy: 60. Result ID: 6a1ebd62fbc373592c19f7a2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### Self-Refine (no early stop) Accuracy: 58. Result ID: 6a1ebd62fbc373592c19f79e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### ATLAS Accuracy: 56. Result ID: 6a1ebd62fbc373592c19f7a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### Self-Refine Accuracy: 53. Result ID: 6a1ebd62fbc373592c19f79d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### GPT-5.2-H Accuracy: 52.48. Result ID: 6a1ebd62fbc373592c19f797. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Protocol":"Zero-shot"}. This record does not assert independent reproduction. ### Reward-model reranking Accuracy: 52. Result ID: 6a1ebd62fbc373592c19f7a0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6","Reward Model":"Skywork-Reward-V2"}. This record does not assert independent reproduction. ### Budget Forcing Accuracy: 51. Result ID: 6a1ebd62fbc373592c19f79f. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. ### Opus-4.6 Accuracy: 50.16. Result ID: 6a1ebd62fbc373592c19f798. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Protocol":"Zero-shot"}. This record does not assert independent reproduction. ### Gemini-3-Pro Accuracy: 48.93. Result ID: 6a1ebd62fbc373592c19f799. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Protocol":"Zero-shot"}. This record does not assert independent reproduction. ### Qwen3-Max-T Accuracy: 48.48. Result ID: 6a1ebd62fbc373592c19f79a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Protocol":"Zero-shot"}. This record does not assert independent reproduction. ### Opus-4.5 Accuracy: 48.16. Result ID: 6a1ebd62fbc373592c19f79b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Protocol":"Zero-shot"}. This record does not assert independent reproduction. ### Pass@1 Accuracy: 48. Result ID: 6a1ebd62fbc373592c19f79c. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Claude Sonnet 4.6"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.