ArgBench (Chain-of-Thought) - Argument Reasoning: 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 reasoning tasks under chain-of-thought prompting, times 100; higher is better.

Metric: Macro F1 (0-1, times 100) averaged over the argument reasoning (fallacies, warrants 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-4.146.5#505
2Mistral-Small-Instruct-240943#2016
3Llama 3.3 70B Instruct39.6#1385
4Llama 3.1 8B Instruct38.9#4600
5Mixtral 8x7B Instruct38.9#4696
6Phi-mini-MoE-instruct38.8
7Qwen 3 32B35.8#972
8Mistral 7B Instruct (v0.3)35.1#6750
9Qwen 3 4B27.8#2348
10DeepSeek R1 Distill Qwen 32B27.4#1834
11DeepSeek-R1-Distill-Qwen-7B27#8055

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-argument-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.