# Maze Navigation on Maze Hard Source: [Maze Navigation on Maze Hard]() Collection index: [llms.txt]() Data updated: 2026-05-20T09:25:50.094Z Evaluation of agent performance on the Maze Hard navigation environment, typically 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 | Maze Navigation | | Dataset (including any stated split) | Maze Hard | | Metric | Accuracy | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 18 of 18 | | 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: SCALAR. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | GPT-5 | 97.66 | Not recorded | 692d2147d6b0dd3342931901 | | OpenAI o3 | 93.36 | Not recorded | 692d2147d6b0dd3342931901 | | PTRM | 86.73 | Not recorded | 6a0d4030f984739ed6737960 | | TRM | 85.3 | Not recorded | 6a0d4030f984739ed6737960 | | Standard TRM, our reproduction | 83.8 | Not recorded | 6a0d4030f984739ed6737960 | | OpenAI o4-mini | 78.52 | Not recorded | 692d2147d6b0dd3342931901 | | HRM | 74.5 | Not recorded | 6a0d4030f984739ed6737960 | | Claude 4.5 Sonnet | 68.36 | Not recorded | 692d2147d6b0dd3342931901 | | Gemini 2.5 Pro | 63.28 | Not recorded | 692d2147d6b0dd3342931901 | | WMAct | 50.59 | Not recorded | 692d2147d6b0dd3342931901 | | PPO - Interactive | 36.52 | Not recorded | 692d2147d6b0dd3342931901 | | Qwen3-14B | 28.52 | Not recorded | 692d2147d6b0dd3342931901 | | PPO - EntirePlan | 26.51 | Not recorded | 692d2147d6b0dd3342931901 | | Qwen3-8B | 17.76 | Not recorded | 692d2147d6b0dd3342931901 | | GPT-4o | 1.56 | Not recorded | 692d2147d6b0dd3342931901 | | Qwen2.5-32B-Instruct | 0.39 | Not recorded | 692d2147d6b0dd3342931901 | | Qwen3-8B-Own | 0.2 | Not recorded | 692d2147d6b0dd3342931901 | | Qwen2.5-7B-Instruct | 0 | Not recorded | 692d2147d6b0dd3342931901 | ### GPT-5 Accuracy: 97.66. Result ID: 699635875698665ec477e7bf. [Reporting paper]() Recorded evaluation configuration: {"category":"Proprietary Models"}. This record does not assert independent reproduction. ### OpenAI o3 Accuracy: 93.36. Result ID: 699635875698665ec477e7c0. [Reporting paper]() Recorded evaluation configuration: {"category":"Proprietary Models"}. This record does not assert independent reproduction. ### PTRM Accuracy: 86.73. Result ID: 6a0d7e0bfbc373592c05d8f1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"# Params":"7M","K (stochastic rollouts)":"100","D (depth)":"16","sigma (exploration parameter)":"1.0","architecture":"TRM-Att"}. This record does not assert independent reproduction. ### TRM Accuracy: 85.3. Result ID: 6a0d7e0bfbc373592c05d8ef. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"# Params":"7M","architecture":"TRM-Att"}. This record does not assert independent reproduction. ### Standard TRM, our reproduction Accuracy: 83.8. Result ID: 6a0d7e0bfbc373592c05d8f0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"# Params":"7M","architecture":"TRM-Att"}. This record does not assert independent reproduction. ### OpenAI o4-mini Accuracy: 78.52. Result ID: 699635875698665ec477e7c1. [Reporting paper]() Recorded evaluation configuration: {"category":"Proprietary Models"}. This record does not assert independent reproduction. ### HRM Accuracy: 74.5. Result ID: 6a0d7e0bfbc373592c05d8ee. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"# Params":"27M"}. This record does not assert independent reproduction. ### Claude 4.5 Sonnet Accuracy: 68.36. Result ID: 699635875698665ec477e7c3. [Reporting paper]() Recorded evaluation configuration: {"category":"Proprietary Models"}. This record does not assert independent reproduction. ### Gemini 2.5 Pro Accuracy: 63.28. Result ID: 699635875698665ec477e7c2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"category":"Proprietary Models"}. This record does not assert independent reproduction. ### WMAct Accuracy: 50.59. Result ID: 699635875698665ec477e7cb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### PPO - Interactive Accuracy: 36.52. Result ID: 699635875698665ec477e7ca. [Reporting paper]() Recorded evaluation configuration: {"mode":"interactive"}. This record does not assert independent reproduction. ### Qwen3-14B Accuracy: 28.52. Result ID: 699635875698665ec477e7c7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"category":"Opensource Models"}. This record does not assert independent reproduction. ### PPO - EntirePlan Accuracy: 26.51. Result ID: 699635875698665ec477e7c9. [Reporting paper]() Recorded evaluation configuration: {"mode":"single-turn output"}. This record does not assert independent reproduction. ### Qwen3-8B Accuracy: 17.76. Result ID: 699635875698665ec477e7c6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"category":"Opensource Models"}. This record does not assert independent reproduction. ### GPT-4o Accuracy: 1.56. Result ID: 699635875698665ec477e7be. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"category":"Proprietary Models"}. This record does not assert independent reproduction. ### Qwen2.5-32B-Instruct Accuracy: 0.39. Result ID: 699635875698665ec477e7c5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"category":"Opensource Models"}. This record does not assert independent reproduction. ### Qwen3-8B-Own Accuracy: 0.2. Result ID: 699635875698665ec477e7c8. [Reporting paper]() Recorded evaluation configuration: {"backbone":"Qwen3-8B"}. This record does not assert independent reproduction. ### Qwen2.5-7B-Instruct Accuracy: 0. Result ID: 699635875698665ec477e7c4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"category":"Opensource Models"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.