ArmBench-LLM - Reading Comprehension: leaderboard
ArmBench-LLM category: Armenian reading comprehension, the mean of five tasks: SQuAD and MS MARCO scored by BLEU (divided by 100 before averaging) and Belebele, DREAM and Hartak public-services multiple choice scored by accuracy. 0-100, higher is better.
Metric: Mean of accuracy and BLEU tasks (0-100). Source: huggingface.co. Saturation forecast: Around April 2028. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 Mini | 62.28 | #228 |
| 2 | GPT-5.2 Pro | 60.96 | #70 |
| 3 | Qwen 3.5 27B | 60.65 | #192 |
| 4 | Grok 4 Fast | 60.37 | #266 |
| 5 | Claude 3.7 Sonnet | 59.66 | #259 |
| 6 | Gemini 3 Pro (Preview) | 59.01 | #71 |
| 7 | Gemini 2.5 Pro | 58.99 | #159 |
| 8 | Gemma 4 31B (IT) | 58.43 | #178 |
| 9 | GLM-4.7 Flash | 58.17 | #543 |
| 10 | GPT-5.4 Nano | 56.47 | #359 |
| 11 | Gemini 3.1 Pro (Preview) | 54.12 | #64 |
| 12 | Claude Sonnet 4 | 53.25 | #207 |
| 13 | Gemini 3 Flash (Preview) | 52.54 | #90 |
| 14 | GLM-5 | 52.52 | #149 |
| 15 | Qwen 3.5 35B A3B | 52.08 | #268 |
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-reading-comprehension · How It Works · Data refreshed daily, snapshot 2026-11-05.