HumanLikeness - Discourse-1: leaderboard
HumanLikeness discourse task comparing model responses with human implicit-causality judgments.
Metric: Humanlike Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Phi-3 Mini 4K Instruct | 79.7 | #1297 |
| 2 | Llama 3.1 70B Instruct | 79.69 | #561 |
| 3 | Llama 3.1 8B Instruct | 79.26 | #1035 |
| 4 | CodeLlama-34B-Instruct-hf | 79.12 | #1381 |
| 5 | Mistral Nemo Instruct (2407) | 78.61 | #936 |
| 6 | Yi 1.5 34B Chat | 77.91 | #800 |
| 7 | c4ai-command-r-plus | 77.7 | #724 |
| 8 | Llama 3 8B Instruct | 77.02 | #1129 |
| 9 | zephyr-7B-alpha | 76.23 | #1312 |
| 10 | GPT-3.5 Turbo | 76.07 | #865 |
| 11 | Llama 2 13B Chat Base | 75.97 | #1422 |
| 12 | Llama 3 70B Instruct | 75.54 | #634 |
| 13 | Llama 2 7B Chat | 75.48 | #1499 |
| 14 | GPT-4o | 75.43 | #343 |
| 15 | starchat2-15B-v0.1 | 75.11 | #832 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=humanlikeness-discourse-1 · How It Works · Data refreshed daily, snapshot 2026-10-20.