LMCA
General QA · 2026-07-29
LMCA (Language Model Conceptual Argumentation) tests how well models judge arguments on conceptual questions in philosophy, decision theory and AI safety. Models rate expert-curated critiques of position texts, and the score is the Pearson correlation between the model's 'overall' ratings and the mean expert 'overall' rating, times 100 and floored at 0. The rated set grows over time (1,461 rated critiques of 560 position texts in October 2026), and the maintainers estimate the ceiling at about 85.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 68.3 |
| Claude Opus 5.5 | 68.2 |
| Claude Fable 5.1 | 65.5 |
| Claude Opus 5 | 64.5 |
| GPT-6 Astra | 64.4 |
| Gemini 4 Argon | 63.7 |
| GPT-6.1 Sol | 62.8 |
| Claude Fable 5 | 61.1 |
| GPT-6 Sol | 59.1 |
| GPT-5.6 Sol | 58.5 |
| Opus 4.8 | 57.5 |
| Muse Wonder | 56.8 |
| Opus 4.6 | 55.8 |
| GLM-5.3 | 55.5 |
| GPT-5.6 Terra | 55.0 |
| GPT-5.5 | 54.3 |
| GPT-5.5 Pro | 53.9 |
| Muse Spark 1.3 | 53.9 |
| Gemini 3.1 Pro Preview | 53.8 |
| Gemini 3.8 Flash | 52.9 |
| Kimi K3 | 52.7 |
| Opus 4.7 | 52.2 |
| GPT-5.4 | 52.0 |
| Claude Haiku 5.5 | 51.1 |
| Gemini 3.7 Flash | 50.4 |
| Sonnet 5 | 50.0 |
| Muse Spark 1.1 | 49.9 |
| Grok 4.7 | 49.4 |
| GPT-5.6 Luna | 48.5 |
| Grok 4.6 | 48.5 |
| Gemini 3.5 Flash | 47.1 |
| DeepSeek V4.1 Flash | 47.0 |
| Sonnet 4.6 | 46.5 |
| Qwen3.8-Max | 46.2 |
| Qwen3.8 2.4T A95B | 45.5 |
| DeepSeek V4 Pro 0813 | 45.5 |
| Gemini 3.6 Flash | 44.9 |
| GPT-6 Luna | 44.5 |
| Opus 4.5 | 44.5 |
| Qwen3.7-Max (source unspecified) | 44.0 |