# Commonsense Reasoning and Knowledge Understanding on ARC, HellaSwag, LAMBADA, PIQA, WinoGrande, and MMLU Source: [Commonsense Reasoning and Knowledge Understanding on ARC, HellaSwag, LAMBADA, PIQA, WinoGrande, and MMLU]() Collection index: [llms.txt]() Data updated: 2026-04-16T07:03:08.035Z Evaluation of model performance across a suite combining core commonsense reasoning tasks (ARC-e, ARC-c, HellaSwag, LAMBADA, PIQA, WinoGrande) and general knowledge understanding (MMLU). 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 | Commonsense Reasoning and Knowledge Understanding | | Dataset (including any stated split) | Commonsense Reasoning and Knowledge Suite (ARC, HellaSwag, LAMBADA, PIQA, WinoGrande, MMLU) | | Metric | ARC-e Accuracy | | 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 - [ARC-e Accuracy](): Higher is better; unit: PERCENT. - [ARC-c Accuracy](): Higher is better; unit: PERCENT. - [HellaSwag Accuracy](): Higher is better; unit: PERCENT. - [LAMBADA Accuracy](): Higher is better; unit: PERCENT. - [PIQA Accuracy](): Higher is better; unit: PERCENT. - [WinoGrande Accuracy](): Higher is better; unit: PERCENT. - [MMLU Accuracy](): Higher is better; unit: PERCENT. - [Average Reasoning Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | ARC-e Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Qwen3-8B | 83.42 | Not recorded | 68fa7c101e209131cea8c392 | | FalconMamba-7B | 83.33 | Not recorded | 68fa7c101e209131cea8c392 | | NHA-Qwen2.5-7B | 82.74 | Not recorded | 68fa7c101e209131cea8c392 | | Mistral-7B | 80.85 | Not recorded | 68fa7c101e209131cea8c392 | | NHA-Llama-3-8B | 80.76 | Not recorded | 68fa7c101e209131cea8c392 | | Qwen2.5-7B | 80.43 | Not recorded | 68fa7c101e209131cea8c392 | | Llama-3-8B | 80.3 | Not recorded | 68fa7c101e209131cea8c392 | | Zamba2-7B | 80.13 | Not recorded | 68fa7c101e209131cea8c392 | | StripedHyena-7B | 78.62 | Not recorded | 68fa7c101e209131cea8c392 | | Mamba-7B | 77.61 | Not recorded | 68fa7c101e209131cea8c392 | | Griffin-7B | 75.4 | Not recorded | 68fa7c101e209131cea8c392 | | RWKV-6-World-7B | 73.61 | Not recorded | 68fa7c101e209131cea8c392 | | Zamba-7B | 71.72 | Not recorded | 68fa7c101e209131cea8c392 | ### Qwen3-8B ARC-e Accuracy: 83.42. Result ID: 69e0898a47f34ce924c94d52. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Transformer","Training Tokens (B)":"36000"}. This record does not assert independent reproduction. ### FalconMamba-7B ARC-e Accuracy: 83.33. Result ID: 69e0898a47f34ce924c94d56. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Linear/Subquadratic","Training Tokens (B)":"5500"}. This record does not assert independent reproduction. ### NHA-Qwen2.5-7B ARC-e Accuracy: 82.74. Result ID: 69e0898a47f34ce924c94d5c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Architecture":"Hybrid","Training Tokens (B)":"10","Full-attention layers":"4"}. This record does not assert independent reproduction. ### Mistral-7B ARC-e Accuracy: 80.85. Result ID: 69e0898a47f34ce924c94d51. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Architecture":"Transformer","Training Tokens (B)":"8000"}. This record does not assert independent reproduction. ### NHA-Llama-3-8B ARC-e Accuracy: 80.76. Result ID: 69e0898a47f34ce924c94d5d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Architecture":"Hybrid","Training Tokens (B)":"10","Full-attention layers":"4"}. This record does not assert independent reproduction. ### Qwen2.5-7B ARC-e Accuracy: 80.43. Result ID: 69e0898a47f34ce924c94d53. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Transformer","Training Tokens (B)":"18000"}. This record does not assert independent reproduction. ### Llama-3-8B ARC-e Accuracy: 80.3. Result ID: 69e0898a47f34ce924c94d54. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Transformer","Training Tokens (B)":"15000"}. This record does not assert independent reproduction. ### Zamba2-7B ARC-e Accuracy: 80.13. Result ID: 69e0898a47f34ce924c94d5b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Architecture":"Hybrid","Training Tokens (B)":"2100","Full-attention layers":"9"}. This record does not assert independent reproduction. ### StripedHyena-7B ARC-e Accuracy: 78.62. Result ID: 69e0898a47f34ce924c94d59. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Hybrid","Full-attention layers":"16"}. This record does not assert independent reproduction. ### Mamba-7B ARC-e Accuracy: 77.61. Result ID: 69e0898a47f34ce924c94d55. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Architecture":"Linear/Subquadratic","Training Tokens (B)":"1200"}. This record does not assert independent reproduction. ### Griffin-7B ARC-e Accuracy: 75.4. Result ID: 69e0898a47f34ce924c94d58. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Linear/Subquadratic","Training Tokens (B)":"300"}. This record does not assert independent reproduction. ### RWKV-6-World-7B ARC-e Accuracy: 73.61. Result ID: 69e0898a47f34ce924c94d57. [Reporting paper]() Recorded evaluation configuration: {"Architecture":"Linear/Subquadratic","Training Tokens (B)":"1420"}. This record does not assert independent reproduction. ### Zamba-7B ARC-e Accuracy: 71.72. Result ID: 69e0898a47f34ce924c94d5a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Architecture":"Hybrid","Training Tokens (B)":"1000","Full-attention layers":"13"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.