# LLM-as-a-Judge Robustness on Sage (Hard) Source: [LLM-as-a-Judge Robustness on Sage (Hard)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:45:08.520Z Evaluation of LLM performance when acting as a judge, specifically assessing robustness on the Sage (Hard) dataset across dimensions like Factuality, Mathematics, and Safety, using IPI and TOV 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 | LLM-as-a-Judge Robustness | | Dataset (including any stated split) | Sage (Hard) | | Metric | Factuality (IPI) | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 13 of 13 | | 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 - [Factuality (IPI)](): Higher is better; unit: PERCENT. - [Factuality (TOV)](): Higher is better; unit: SCALAR. - [Precise IF (IPI)](): Higher is better; unit: PERCENT. - [Precise IF (TOV)](): Higher is better; unit: SCALAR. - [Mathematics (IPI)](): Higher is better; unit: PERCENT. - [Mathematics (TOV)](): Higher is better; unit: SCALAR. - [Safety (IPI)](): Higher is better; unit: PERCENT. - [Safety (TOV)](): Higher is better; unit: SCALAR. - [Focus (IPI)](): Higher is better; unit: PERCENT. - [Focus (TOV)](): Higher is better; unit: SCALAR. - [Overall (IPI)](): Higher is better; unit: PERCENT. - [Overall (TOV)](): Higher is better; unit: SCALAR. ## Recorded results | Method | Factuality (IPI) | Result date | Report paper | | --- | --- | --- | --- | | Llama-3.1-8B-Instruct | 55.9 | Not recorded | 6944dbe0423ea422657298b3 | | GPT-4o-mini | 54.6 | Not recorded | 6944dbe0423ea422657298b3 | | Claude-3-Haiku | 52.2 | Not recorded | 6944dbe0423ea422657298b3 | | DeepSeek-V3.1 | 51.3 | Not recorded | 6944dbe0423ea422657298b3 | | DeepSeek-V3-0324 | 45.5 | Not recorded | 6944dbe0423ea422657298b3 | | Gemini-2.0-Flash-Lite | 43.4 | Not recorded | 6944dbe0423ea422657298b3 | | GPT-5-Chat | 43.3 | Not recorded | 6944dbe0423ea422657298b3 | | Qwen3-30B-A3B-Instruct-2507 | 42.4 | Not recorded | 6944dbe0423ea422657298b3 | | DeepSeek-R1-0528 | 41 | Not recorded | 6944dbe0423ea422657298b3 | | Qwen3-4B-Instruct-2507 | 37.6 | Not recorded | 6944dbe0423ea422657298b3 | | Qwen3-235B-A22B-Instruct-2507 | 34.7 | Not recorded | 6944dbe0423ea422657298b3 | | Gemini-2.5-Flash | 26.5 | Not recorded | 6944dbe0423ea422657298b3 | | Gemini-2.5-Pro | 24 | Not recorded | 6944dbe0423ea422657298b3 | ### Llama-3.1-8B-Instruct Factuality (IPI): 55.9. Result ID: 6996bb635698665ec4062d59. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### GPT-4o-mini Factuality (IPI): 54.6. Result ID: 6996bb635698665ec4062d58. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Claude-3-Haiku Factuality (IPI): 52.2. Result ID: 6996bb635698665ec4062d56. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### DeepSeek-V3.1 Factuality (IPI): 51.3. Result ID: 6996bb635698665ec4062d57. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### DeepSeek-V3-0324 Factuality (IPI): 45.5. Result ID: 6996bb635698665ec4062d55. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Gemini-2.0-Flash-Lite Factuality (IPI): 43.4. Result ID: 6996bb635698665ec4062d53. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### GPT-5-Chat Factuality (IPI): 43.3. Result ID: 6996bb635698665ec4062d54. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Qwen3-30B-A3B-Instruct-2507 Factuality (IPI): 42.4. Result ID: 6996bb635698665ec4062d52. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### DeepSeek-R1-0528 Factuality (IPI): 41. Result ID: 6996bb635698665ec4062d51. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Qwen3-4B-Instruct-2507 Factuality (IPI): 37.6. Result ID: 6996bb635698665ec4062d50. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Qwen3-235B-A22B-Instruct-2507 Factuality (IPI): 34.7. Result ID: 6996bb635698665ec4062d4f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Gemini-2.5-Flash Factuality (IPI): 26.5. Result ID: 6996bb635698665ec4062d4e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Gemini-2.5-Pro Factuality (IPI): 24. Result ID: 6996bb635698665ec4062d4d. [Reporting paper]() [Method origin paper]() 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.