ArgBench (Chain-of-Thought) - Quality Assessment: leaderboard
ArgBench (2026) unifies 46 computational-argumentation tasks from 33 existing datasets into five skills. This board is the macro F1 averaged over the argument quality assessment tasks under chain-of-thought prompting, times 100; higher is better.
Metric: Macro F1 (0-1, times 100) averaged over the argument quality assessment tasks, chain-of-thought prompting; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct | 43.9 | #1385 |
| 2 | GPT-4.1 | 43.6 | #505 |
| 3 | Qwen 3 32B | 43 | #972 |
| 4 | Qwen 3 4B | 40.7 | #2348 |
| 5 | Mistral 7B Instruct (v0.3) | 40.4 | #6750 |
| 6 | DeepSeek R1 Distill Qwen 32B | 39.7 | #1834 |
| 7 | DeepSeek-R1-Distill-Qwen-7B | 38.9 | #8055 |
| 8 | Llama 3.1 8B Instruct | 36.7 | #4600 |
| 9 | Phi-mini-MoE-instruct | 35.8 | |
| 10 | Mixtral 8x7B Instruct | 35.6 | #4696 |
| 11 | Mistral-Small-Instruct-2409 | 35.3 | #2016 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=argbench-chain-of-thought-quality-assessment · How It Works · Data refreshed daily, snapshot 2026-11-21.