# Mathematical Reasoning on AIME-25 (pass@4) Source: [Mathematical Reasoning on AIME-25 (pass@4)]() Collection index: [llms.txt]() Data updated: 2026-05-20T07:03:31.772Z Evaluation of model performance on the American Invitational Mathematics Examination (AIME) 25 dataset for mathematical reasoning, specifically measured using the pass@4 metric. 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) | AIME-25 | | Metric | Pass@4 | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 25 of 25 | | 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 - [Pass@4](): Higher is better; unit: PERCENT. ## Recorded results | Method | Pass@4 | Result date | Report paper | | --- | --- | --- | --- | | Hybrid-LoRA | 55.3 | Not recorded | 6a0d4012f984739ed67377be | | AutoLoRA | 52.7 | Not recorded | 6a0d4012f984739ed67377be | | Hybrid-LoRA | 52 | Not recorded | 6a0d4012f984739ed67377be | | AutoLoRA | 48.7 | Not recorded | 6a0d4012f984739ed67377be | | Full-FT (ref.) | 47.3 | Not recorded | 6a0d4012f984739ed67377be | | Hybrid-LoRA | 45.3 | Not recorded | 6a0d4012f984739ed67377be | | Full-FT (ref.) | 44 | Not recorded | 6a0d4012f984739ed67377be | | Hybrid-LoRA | 42 | Not recorded | 6a0d4012f984739ed67377be | | AutoLoRA | 40.7 | Not recorded | 6a0d4012f984739ed67377be | | AutoLoRA | 38.7 | Not recorded | 6a0d4012f984739ed67377be | | Full-FT (ref.) | 37.3 | Not recorded | 6a0d4012f984739ed67377be | | Full-FT (ref.) | 36.7 | Not recorded | 6a0d4012f984739ed67377be | | Hybrid-LoRA | 36.7 | Not recorded | 6a0d4012f984739ed67377be | | Hybrid-LoRA | 35.3 | Not recorded | 6a0d4012f984739ed67377be | | LoRA-Drop | 34 | Not recorded | 6a0d4012f984739ed67377be | | AdaLoRA | 33.3 | Not recorded | 6a0d4012f984739ed67377be | | AutoLoRA | 32.7 | Not recorded | 6a0d4012f984739ed67377be | | LoRA-Drop | 32 | Not recorded | 6a0d4012f984739ed67377be | | AutoLoRA | 31.3 | Not recorded | 6a0d4012f984739ed67377be | | AdaLoRA | 31.3 | Not recorded | 6a0d4012f984739ed67377be | | LoRA-FFN | 30.7 | Not recorded | 6a0d4012f984739ed67377be | | Full-FT | 21.3 | Not recorded | 6a0d4012f984739ed67377be | | CoT | 20.6 | Not recorded | 6a0d4012f984739ed67377be | | COT | 10.7 | Not recorded | 6a0d4012f984739ed67377be | | IO | 0 | Not recorded | 6a0d4012f984739ed67377be | ### Hybrid-LoRA Pass@4: 55.3. Result ID: 6a0d5c4efbc373592c04e6bf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 7B Instruct","Post-training Approach":"GSPO","#Tunable Params":"449M"}. This record does not assert independent reproduction. ### AutoLoRA Pass@4: 52.7. Result ID: 6a0d5c4efbc373592c04e6be. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 7B Instruct","Post-training Approach":"GSPO","#Tunable Params":"453M"}. This record does not assert independent reproduction. ### Hybrid-LoRA Pass@4: 52. Result ID: 6a0d5c4efbc373592c04e6bd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 7B Instruct","Post-training Approach":"GRPO","#Tunable Params":"449M"}. This record does not assert independent reproduction. ### AutoLoRA Pass@4: 48.7. Result ID: 6a0d5c4efbc373592c04e6bc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 7B Instruct","Post-training Approach":"GRPO","#Tunable Params":"453M"}. This record does not assert independent reproduction. ### Full-FT (ref.) Pass@4: 47.3. Result ID: 6a0d5c4efbc373592c04e6b9. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"GSPO","#Tunable Params":"3086M"}. This record does not assert independent reproduction. ### Hybrid-LoRA Pass@4: 45.3. Result ID: 6a0d5c4efbc373592c04e6bb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"GSPO","#Tunable Params":"192M"}. This record does not assert independent reproduction. ### Full-FT (ref.) Pass@4: 44. Result ID: 6a0d5c4efbc373592c04e6b6. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"GRPO","#Tunable Params":"3086M"}. This record does not assert independent reproduction. ### Hybrid-LoRA Pass@4: 42. Result ID: 6a0d5c4efbc373592c04e6b8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"GRPO","#Tunable Params":"192M"}. This record does not assert independent reproduction. ### AutoLoRA Pass@4: 40.7. Result ID: 6a0d5c4efbc373592c04e6ba. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"GSPO","#Tunable Params":"194M"}. This record does not assert independent reproduction. ### AutoLoRA Pass@4: 38.7. Result ID: 6a0d5c4efbc373592c04e6b7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"GRPO","#Tunable Params":"201M"}. This record does not assert independent reproduction. ### Full-FT (ref.) Pass@4: 37.3. Result ID: 6a0d5c4efbc373592c04e6af. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GSPO","#Tunable Params":"1544M"}. This record does not assert independent reproduction. ### Full-FT (ref.) Pass@4: 36.7. Result ID: 6a0d5c4efbc373592c04e6aa. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GRPO","#Tunable Params":"1544M"}. This record does not assert independent reproduction. ### Hybrid-LoRA Pass@4: 36.7. Result ID: 6a0d5c4efbc373592c04e6b4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GSPO","#Tunable Params":"105M"}. This record does not assert independent reproduction. ### Hybrid-LoRA Pass@4: 35.3. Result ID: 6a0d5c4efbc373592c04e6ae. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GRPO","#Tunable Params":"105M"}. This record does not assert independent reproduction. ### LoRA-Drop Pass@4: 34. Result ID: 6a0d5c4efbc373592c04e6b3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GSPO","#Tunable Params":"105M"}. This record does not assert independent reproduction. ### AdaLoRA Pass@4: 33.3. Result ID: 6a0d5c4efbc373592c04e6ab. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GRPO","#Tunable Params":"105M"}. This record does not assert independent reproduction. ### AutoLoRA Pass@4: 32.7. Result ID: 6a0d5c4efbc373592c04e6b2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GSPO","#Tunable Params":"107M"}. This record does not assert independent reproduction. ### LoRA-Drop Pass@4: 32. Result ID: 6a0d5c4efbc373592c04e6ad. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GRPO","#Tunable Params":"105M"}. This record does not assert independent reproduction. ### AutoLoRA Pass@4: 31.3. Result ID: 6a0d5c4efbc373592c04e6ac. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GRPO","#Tunable Params":"106M"}. This record does not assert independent reproduction. ### AdaLoRA Pass@4: 31.3. Result ID: 6a0d5c4efbc373592c04e6b1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GSPO","#Tunable Params":"106M"}. This record does not assert independent reproduction. ### LoRA-FFN Pass@4: 30.7. Result ID: 6a0d5c4efbc373592c04e6b0. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"GSPO","#Tunable Params":"122M"}. This record does not assert independent reproduction. ### Full-FT Pass@4: 21.3. Result ID: 6a0d5c4efbc373592c04e6a9. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"SFT","#Tunable Params":"1544M"}. This record does not assert independent reproduction. ### CoT Pass@4: 20.6. Result ID: 6a0d5c4efbc373592c04e6b5. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 3B Instruct","Post-training Approach":"-","#Tunable Params":"0.0"}. This record does not assert independent reproduction. ### COT Pass@4: 10.7. Result ID: 6a0d5c4efbc373592c04e6a8. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"COT","#Tunable Params":"0.0"}. This record does not assert independent reproduction. ### IO Pass@4: 0. Result ID: 6a0d5c4efbc373592c04e6a7. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-2.5 1.5B Instruct","Post-training Approach":"IO","#Tunable Params":"0.0"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.