# Professional deep-research writing on Deepresearch-Gym Source: [Professional deep-research writing on Deepresearch-Gym]() Collection index: [llms.txt]() Data updated: 2026-04-06T08:44:15.122Z Evaluation of model performance on professional deep-research writing tasks using the Deepresearch-Gym dataset, measured by Key Performance Ratios (KPR, KPC) and overall Quality 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 | Professional deep-research writing | | Dataset (including any stated split) | Deepresearch-Gym | | Metric | KPR | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 19 of 19 | | 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 - [KPR](): Higher is better; unit: SCALAR. - [KPC](): Higher is better; unit: SCALAR. - [Quality](): Higher is better; unit: SCALAR. ## Recorded results | Method | KPR | Result date | Report paper | | --- | --- | --- | --- | | Claude-4-sonnet | 77.1 | Not recorded | 69d33ded7d0172140ec735d6 | | GPT-5 | 76.26 | Not recorded | 69d33ded7d0172140ec735d6 | | Gemini2.5-Pro | 73.2 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + R2-Write-SFT + RLp | 72.5 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + R2-Write-SFT + RL | 71.63 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + R2-Write-SFT | 70.05 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-4B + R2-Write-SFT + RLp | 69.72 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + Distill + RL | 68.92 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + RL | 68.84 | Not recorded | 69d33ded7d0172140ec735d6 | | LongWriter-Zero-32B | 68.2 | Not recorded | 69d33ded7d0172140ec735d6 | | Writing-RL-7B | 68.1 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-4B + R2-Write-SFT | 67.12 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + Distill | 66.37 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B + R2-Write-Last | 65.72 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-4B + RL | 64.2 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-8B | 63.78 | Not recorded | 69d33ded7d0172140ec735d6 | | Qwen3-4B | 60.04 | Not recorded | 69d33ded7d0172140ec735d6 | | Reverse-Engineering-8B | 60.01 | Not recorded | 69d33ded7d0172140ec735d6 | | Longwriter-9B | 57.22 | Not recorded | 69d33ded7d0172140ec735d6 | ### Claude-4-sonnet KPR: 77.1. Result ID: 69d3725e47f34ce924bb5c84. [Reporting paper]() Recorded evaluation configuration: {"Training":"-"}. This record does not assert independent reproduction. ### GPT-5 KPR: 76.26. Result ID: 69d3725e47f34ce924bb5c82. [Reporting paper]() Recorded evaluation configuration: {"Training":"-"}. This record does not assert independent reproduction. ### Gemini2.5-Pro KPR: 73.2. Result ID: 69d3725e47f34ce924bb5c83. [Reporting paper]() Recorded evaluation configuration: {"Training":"-"}. This record does not assert independent reproduction. ### Qwen3-8B + R2-Write-SFT + RLp KPR: 72.5. Result ID: 69d3725e47f34ce924bb5c94. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT + PPO"}. This record does not assert independent reproduction. ### Qwen3-8B + R2-Write-SFT + RL KPR: 71.63. Result ID: 69d3725e47f34ce924bb5c93. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT + PPO"}. This record does not assert independent reproduction. ### Qwen3-8B + R2-Write-SFT KPR: 70.05. Result ID: 69d3725e47f34ce924bb5c8c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT"}. This record does not assert independent reproduction. ### Qwen3-4B + R2-Write-SFT + RLp KPR: 69.72. Result ID: 69d3725e47f34ce924bb5c92. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT + PPO"}. This record does not assert independent reproduction. ### Qwen3-8B + Distill + RL KPR: 68.92. Result ID: 69d3725e47f34ce924bb5c8f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT + PPO"}. This record does not assert independent reproduction. ### Qwen3-8B + RL KPR: 68.84. Result ID: 69d3725e47f34ce924bb5c8e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"PPO"}. This record does not assert independent reproduction. ### LongWriter-Zero-32B KPR: 68.2. Result ID: 69d3725e47f34ce924bb5c91. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT + GRPO"}. This record does not assert independent reproduction. ### Writing-RL-7B KPR: 68.1. Result ID: 69d3725e47f34ce924bb5c90. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT + PPO"}. This record does not assert independent reproduction. ### Qwen3-4B + R2-Write-SFT KPR: 67.12. Result ID: 69d3725e47f34ce924bb5c8b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT"}. This record does not assert independent reproduction. ### Qwen3-8B + Distill KPR: 66.37. Result ID: 69d3725e47f34ce924bb5c89. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT"}. This record does not assert independent reproduction. ### Qwen3-8B + R2-Write-Last KPR: 65.72. Result ID: 69d3725e47f34ce924bb5c8a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT"}. This record does not assert independent reproduction. ### Qwen3-4B + RL KPR: 64.2. Result ID: 69d3725e47f34ce924bb5c8d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"PPO"}. This record does not assert independent reproduction. ### Qwen3-8B KPR: 63.78. Result ID: 69d3725e47f34ce924bb5c86. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"-"}. This record does not assert independent reproduction. ### Qwen3-4B KPR: 60.04. Result ID: 69d3725e47f34ce924bb5c85. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"-"}. This record does not assert independent reproduction. ### Reverse-Engineering-8B KPR: 60.01. Result ID: 69d3725e47f34ce924bb5c88. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT"}. This record does not assert independent reproduction. ### Longwriter-9B KPR: 57.22. Result ID: 69d3725e47f34ce924bb5c87. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training":"SFT"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.