ArmBench-LLM - Overall: leaderboard
ArmBench-LLM (Metric AI Lab, Yerevan): Armenian-language benchmark run with Metric's lighteval fork. Unweighted mean of ten category scores (NER, POS, reading comprehension, classification, MCQA, generation, translation, national exams rescaled from 0-20, text processing, MMLU-Pro), 0-100, higher is better. The site excludes Claude 4.5 and 4.6 models, citing evaluation reliability issues.
Metric: Mean category score (0-100; exams rescaled from 0-20). Source: huggingface.co. Saturation forecast: Around November 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 63.5 | #90 |
| 2 | GPT-5.2 Pro | 61.71 | #70 |
| 3 | Gemma 4 31B (IT) | 61.42 | #178 |
| 4 | Claude 3.7 Sonnet | 60.71 | #259 |
| 5 | Claude Sonnet 4 | 60.37 | #207 |
| 6 | Grok 4 Fast | 60.37 | #266 |
| 7 | Gemini 3 Pro (Preview) | 59.49 | #71 |
| 8 | Gemini 2.5 Flash | 58.19 | #255 |
| 9 | Grok 4.1 Fast | 57.68 | #227 |
| 10 | Qwen 3.5 27B | 57.67 | #192 |
| 11 | GPT-5.4 Mini | 56.89 | #228 |
| 12 | Gemini 3.1 Pro (Preview) | 52.2 | #64 |
| 13 | GLM-5 | 51.57 | #149 |
| 14 | Gemini 2.5 Pro | 50.36 | #159 |
| 15 | Mistral Large 3 | 50.04 | #410 |
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-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.