HumanLikeness - Discourse-2: leaderboard
HumanLikeness Discourse-2 — how human-like model text is at the discourse level, judged against human writing; humanlike score.
Metric: Humanlike Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 2 13B Chat Base | 78.85 | #1422 |
| 2 | Phi-3 Mini 4K Instruct | 76.18 | #1297 |
| 3 | zephyr-7B-alpha | 74.58 | #1312 |
| 4 | Mistral Nemo Instruct (2407) | 67.84 | #936 |
| 5 | Llama 3.1 70B Instruct | 63.12 | #561 |
| 6 | GPT-4o | 62.11 | #343 |
| 7 | starchat2-15B-v0.1 | 61.81 | #832 |
| 8 | CodeLlama-34B-Instruct-hf | 60.74 | #1381 |
| 9 | zephyr-7B-beta | 59.56 | #1330 |
| 10 | GPT-4o Mini | 57.32 | #599 |
| 11 | Llama 3.1 8B Instruct | 55.57 | #1035 |
| 12 | Llama 3 8B Instruct | 54.28 | #1129 |
| 13 | Mixtral 8x7B Instruct (v0.1) | 52.18 | #1073 |
| 14 | c4ai-command-r-plus | 50.44 | #724 |
| 15 | Llama 3 70B Instruct | 50.4 | #634 |
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-2 · How It Works · Data refreshed daily, snapshot 2026-10-20.