ArmBench-LLM - MMLU-Pro: leaderboard
ArmBench-LLM category: MMLU-Pro in Armenian, multiple-choice accuracy across its academic and professional subjects, evaluated with Metric AI Lab's lighteval fork. 0-100, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 86.79 | #90 |
| 2 | GPT-5.2 Pro | 84.88 | #70 |
| 3 | Grok 4.1 Fast | 81.68 | #227 |
| 4 | Grok 4 Fast | 81.68 | #266 |
| 5 | Qwen 3.5 27B | 81.28 | #192 |
| 6 | Gemini 3 Pro (Preview) | 80.78 | #71 |
| 7 | Gemma 4 31B (IT) | 79.48 | #178 |
| 8 | Gemini 3.1 Pro (Preview) | 75.08 | #64 |
| 9 | Claude Sonnet 4 | 74.77 | #207 |
| 10 | Claude 3.7 Sonnet | 74.57 | #259 |
| 11 | Mistral Large 3 | 74.07 | #410 |
| 12 | GPT-5.4 Mini | 70.47 | #228 |
| 13 | GLM-5 | 69.47 | #149 |
| 14 | Qwen 3.5 35B A3B | 65.27 | #268 |
| 15 | Gemini 2.5 Flash | 64.16 | #255 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=armbench-llm-mmlu-pro · How It Works · Data refreshed daily, snapshot 2026-11-05.