ArgBench (Chain-of-Thought) - Perspective 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 perspective assessment tasks under chain-of-thought prompting, times 100; higher is better.

Metric: Macro F1 (0-1, times 100) averaged over the argument perspective assessment (stance, framing and similar) 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

#ModelScoreOverall rank
1GPT-4.160.3#505
2Qwen 3 32B55.2#972
3Llama 3.3 70B Instruct53.1#1385
4Mixtral 8x7B Instruct52.3#4696
5Llama 3.1 8B Instruct49.2#4600
6Qwen 3 4B48.4#2348
7DeepSeek R1 Distill Qwen 32B47.8#1834
8Mistral-Small-Instruct-240947.4#2016
9DeepSeek-R1-Distill-Qwen-7B42.2#8055
10Phi-mini-MoE-instruct39.8
11Mistral 7B Instruct (v0.3)38.8#6750

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-perspective-assessment · How It Works · Data refreshed daily, snapshot 2026-11-21.