# Mathematical Reasoning on GSM Hard (Accuracy) Source: [Mathematical Reasoning on GSM Hard (Accuracy)]() Collection index: [llms.txt]() Data updated: 2026-05-18T07:38:08.757Z Evaluation of model performance on challenging mathematical reasoning problems from the GSM Hard dataset, reporting only the Accuracy score. 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 | Mathematical Reasoning | | Dataset (including any stated split) | GSM Hard | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 31 of 31 | | 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 | | --- | --- | --- | --- | | RM-Regen | 68.6 | Not recorded | 69c21a8e0636e5bf4afc4895 | | RM-Regen | 64 | Not recorded | 69c21a8e0636e5bf4afc4895 | | RM-Regen | 53 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Reflexion(3 iters) | 50.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | RM-Regen | 49.4 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ProCo | 48.6 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Best-of-N (N=3) | 48.4 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Best-of-N (N=3) | 46.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ST CoT | 46.2 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Reflexion(3 iters) | 45.2 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ReflectEvo | 44.4 | Not recorded | 69c21a8e0636e5bf4afc4895 | | RM-Regen | 41 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ST CoT | 39.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ProCo | 39.6 | Not recorded | 69c21a8e0636e5bf4afc4895 | | NEXA | 37.13 | Not recorded | 6a0a9d1edbc836b87d82b47f | | Reflexion(3 iters) | 36.2 | Not recorded | 69c21a8e0636e5bf4afc4895 | | CoT | 36.07 | Not recorded | 6a0a9d1edbc836b87d82b47f | | RM-Regen | 34.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Single | 34.4 | Not recorded | 6a0a9d1edbc836b87d82b47f | | Best-of-N (N=3) | 34.2 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Self-Refine | 32.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | Self-Refine | 32.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ReflectEvo | 32 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ST CoT | 31.8 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ProCo | 31.2 | Not recorded | 69c21a8e0636e5bf4afc4895 | | ALIGN | 24.6 | Not recorded | 69818127a8b01e76cc8e22f9 | | SC | 24.1 | Not recorded | 69818127a8b01e76cc8e22f9 | | Self-Refine | 23 | Not recorded | 69c21a8e0636e5bf4afc4895 | | rStar | 22.8 | Not recorded | 69818127a8b01e76cc8e22f9 | | FS-CoT | 17.7 | Not recorded | 69818127a8b01e76cc8e22f9 | | Principal | 15.6 | Not recorded | 69818127a8b01e76cc8e22f9 | ### RM-Regen Accuracy: 68.6. Result ID: 69c2adfa47f34ce924abc1a0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-3.5","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### RM-Regen Accuracy: 64. Result ID: 69c2adfa47f34ce924abc20f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-3.5"}. This record does not assert independent reproduction. ### RM-Regen Accuracy: 53. Result ID: 69c2adfa47f34ce924abc1a8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### Reflexion(3 iters) Accuracy: 50.8. Result ID: 69c2adfa47f34ce924abc1a5. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemma2-9B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### RM-Regen Accuracy: 49.4. Result ID: 69c2adfa47f34ce924abc217. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B"}. This record does not assert independent reproduction. ### ProCo Accuracy: 48.6. Result ID: 69c2adfa47f34ce924abc215. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B","iterations":"3"}. This record does not assert independent reproduction. ### Best-of-N (N=3) Accuracy: 48.4. Result ID: 69c2adfa47f34ce924abc1a7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### Best-of-N (N=3) Accuracy: 46.8. Result ID: 69c2adfa47f34ce924abc19f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-3.5","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### ST CoT Accuracy: 46.2. Result ID: 69c2adfa47f34ce924abc216. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B","iterations":"3"}. This record does not assert independent reproduction. ### Reflexion(3 iters) Accuracy: 45.2. Result ID: 69c2adfa47f34ce924abc19e. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"GPT-3.5","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### ReflectEvo Accuracy: 44.4. Result ID: 69c2adfa47f34ce924abc1a6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### RM-Regen Accuracy: 41. Result ID: 69c2adfa47f34ce924abc1a4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### ST CoT Accuracy: 39.8. Result ID: 69c2adfa47f34ce924abc20e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-3.5","iterations":"3"}. This record does not assert independent reproduction. ### ProCo Accuracy: 39.6. Result ID: 69c2adfa47f34ce924abc20d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-3.5","iterations":"3"}. This record does not assert independent reproduction. ### NEXA Accuracy: 37.13. Result ID: 6a0ac1c547f34ce924f08b6f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Trained on":"GSM8K","Evaluation Protocol":"Without retraining"}. This record does not assert independent reproduction. ### Reflexion(3 iters) Accuracy: 36.2. Result ID: 69c2adfa47f34ce924abc1a1. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### CoT Accuracy: 36.07. Result ID: 6a0ac1c547f34ce924f08b6e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Trained on":"GSM8K","Evaluation Protocol":"Without retraining"}. This record does not assert independent reproduction. ### RM-Regen Accuracy: 34.8. Result ID: 69c2adfa47f34ce924abc213. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B"}. This record does not assert independent reproduction. ### Single Accuracy: 34.4. Result ID: 6a0ac1c547f34ce924f08b6d. [Reporting paper]() Recorded evaluation configuration: {"Trained on":"GSM8K","Evaluation Protocol":"Without retraining"}. This record does not assert independent reproduction. ### Best-of-N (N=3) Accuracy: 34.2. Result ID: 69c2adfa47f34ce924abc1a3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### Self-Refine Accuracy: 32.8. Result ID: 69c2adfa47f34ce924abc20c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"GPT-3.5","iterations":"3"}. This record does not assert independent reproduction. ### Self-Refine Accuracy: 32.8. Result ID: 69c2adfa47f34ce924abc214. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma2-9B","iterations":"3"}. This record does not assert independent reproduction. ### ReflectEvo Accuracy: 32. Result ID: 69c2adfa47f34ce924abc1a2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","Verification Setting":"Oracle"}. This record does not assert independent reproduction. ### ST CoT Accuracy: 31.8. Result ID: 69c2adfa47f34ce924abc212. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","iterations":"3"}. This record does not assert independent reproduction. ### ProCo Accuracy: 31.2. Result ID: 69c2adfa47f34ce924abc211. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","iterations":"3"}. This record does not assert independent reproduction. ### ALIGN Accuracy: 24.6. Result ID: 699909b75698665ec46f37fc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Reasoning Strategy":"ALIGN"}. This record does not assert independent reproduction. ### SC Accuracy: 24.1. Result ID: 699909b75698665ec46f37f9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Reasoning Strategy":"CoT-SC@maj16"}. This record does not assert independent reproduction. ### Self-Refine Accuracy: 23. Result ID: 69c2adfa47f34ce924abc210. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama3.1-8B","iterations":"3"}. This record does not assert independent reproduction. ### rStar Accuracy: 22.8. Result ID: 699909b75698665ec46f37fb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Reasoning Strategy":"rStar"}. This record does not assert independent reproduction. ### FS-CoT Accuracy: 17.7. Result ID: 699909b75698665ec46f37f8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Reasoning Strategy":"Few-shot CoT"}. This record does not assert independent reproduction. ### Principal Accuracy: 15.6. Result ID: 699909b75698665ec46f37fa. [Reporting paper]() Recorded evaluation configuration: {"Reasoning Strategy":"Principal"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.