Scientific Manuscript Reviewing on ICLR 2026 (test)
0.55Actionability ScoreProReviewer
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ProReviewerBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.55 | 0.76 | 0.57 | 0.48 | 0.9 | 0.65 | |
| CycleReviewerBackbone=Llama3.1-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.49 | 0.34 | 0.33 | 0.23 | 0.87 | 0.45 | |
| Vanilla RLBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.49 | 0.61 | 0.47 | 0.43 | 0.89 | 0.58 | |
| DeepReviewBackbone=Llama3.1-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.47 | 0.5 | 0.36 | 0.33 | 0.92 | 0.51 | |
| DeepReviewBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.46 | 0.33 | 0.32 | 0.23 | 0.9 | 0.45 | |
| ProReviewerBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.46 | 0.64 | 0.48 | 0.4 | 0.88 | 0.57 | |
| ProReviewerBackbone=Llama3.1-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.4 | 0.7 | 0.54 | 0.5 | 0.9 | 0.61 | |
| CycleReviewerBackbone=Llama3.1-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.39 | 0.25 | 0.26 | 0.15 | 0.84 | 0.38 | |
| AgentReviewBackbone=Gemini-3.1-flash-lite, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.37 | 0.46 | 0.5 | 0.33 | 0.77 | 0.49 | |
| CycleReviewerBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Best-of-42026.06 | 0.37 | 0.36 | 0.27 | 0.21 | 0.84 | 0.41 | |
| AI-Scientist-v2Backbone=Qwen3.5-397B-A17B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.36 | 0.55 | 0.52 | 0.31 | 0.83 | 0.51 | |
| AgentReviewBackbone=Qwen3-8B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.35 | 0.4 | 0.26 | 0.18 | 0.66 | 0.37 | |
| AgentReviewBackbone=Qwen3.5-397B-A17B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.35 | 0.57 | 0.52 | 0.29 | 0.79 | 0.5 | |
| DeepReviewBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.35 | 0.22 | 0.23 | 0.13 | 0.85 | 0.35 | |
| AI-Scientist-v2Backbone=Gemini-3.1-flash-lite, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.34 | 0.41 | 0.47 | 0.26 | 0.81 | 0.46 | |
| DeepReviewBackbone=Llama3.1-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.34 | 0.35 | 0.26 | 0.24 | 0.85 | 0.41 | |
| Vanilla RLBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.33 | 0.51 | 0.36 | 0.38 | 0.89 | 0.49 | |
| CycleReviewerBackbone=Qwen3-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.31 | 0.28 | 0.21 | 0.17 | 0.84 | 0.36 | |
| AgentReviewBackbone=Qwen3-8B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.3 | 0.32 | 0.21 | 0.13 | 0.63 | 0.32 | |
| ProReviewerBackbone=Llama3.1-8B, Training Type=Fine-tuned, Sampling Protocol=Avg-of-42026.06 | 0.3 | 0.56 | 0.44 | 0.4 | 0.89 | 0.52 | |
| AgentReviewBackbone=Gemini-3.1-flash-lite, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.29 | 0.36 | 0.44 | 0.25 | 0.7 | 0.41 | |
| AgentReviewBackbone=Qwen3.5-397B-A17B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.28 | 0.46 | 0.45 | 0.23 | 0.78 | 0.44 | |
| AI-Scientist-v2Backbone=Qwen3.5-397B-A17B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.28 | 0.44 | 0.46 | 0.25 | 0.83 | 0.45 | |
| AI-Scientist-v2Backbone=Qwen3-14B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.27 | 0.29 | 0.28 | 0.15 | 0.73 | 0.34 | |
| AI-Scientist-v2Backbone=Qwen3-8B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.26 | 0.28 | 0.27 | 0.12 | 0.69 | 0.32 | |
| AI-Scientist-v2Backbone=Qwen3-32B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.26 | 0.33 | 0.35 | 0.2 | 0.73 | 0.37 | |
| AI-Scientist-v2Backbone=Gemini-3.1-flash-lite, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.26 | 0.31 | 0.38 | 0.19 | 0.73 | 0.37 | |
| AI-Scientist-v2Backbone=Llama3.1-8B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.25 | 0.26 | 0.2 | 0.09 | 0.81 | 0.32 | |
| AgentReviewBackbone=Qwen3-32B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.24 | 0.32 | 0.38 | 0.22 | 0.69 | 0.37 | |
| AgentReviewBackbone=Qwen3-14B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.22 | 0.26 | 0.3 | 0.15 | 0.63 | 0.31 | |
| AI-Scientist-v2Backbone=Qwen3-8B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.21 | 0.21 | 0.21 | 0.09 | 0.68 | 0.28 | |
| AI-Scientist-v2Backbone=Qwen3-14B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.21 | 0.21 | 0.22 | 0.11 | 0.69 | 0.29 | |
| AI-Scientist-v2Backbone=Qwen3-32B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.21 | 0.24 | 0.28 | 0.14 | 0.69 | 0.31 | |
| AgentReviewBackbone=Qwen3-32B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.2 | 0.25 | 0.31 | 0.16 | 0.64 | 0.31 | |
| AgentReviewBackbone=Qwen3-14B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.19 | 0.2 | 0.24 | 0.11 | 0.61 | 0.27 | |
| AI-Scientist-v2Backbone=Llama3.1-8B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.17 | 0.16 | 0.14 | 0.07 | 0.74 | 0.25 | |
| AgentReviewBackbone=Llama3.1-8B, Training Type=Prompt-based, Sampling Protocol=Best-of-42026.06 | 0.14 | 0.13 | 0.14 | 0.07 | 0.66 | 0.23 | |
| AgentReviewBackbone=Llama3.1-8B, Training Type=Prompt-based, Sampling Protocol=Avg-of-42026.06 | 0.13 | 0.12 | 0.13 | 0.06 | 0.64 | 0.21 |