ArmBench-LLM v2 - Reading: leaderboard

ArmBench-LLM 5-category edition Reading score: mean across Belebele, Syndarin, Scientific, DREAM multiple-choice accuracy and SQuAD and MS MARCO (BLEU/100) in Armenian. 0-100, higher is better.

Metric: Mean of accuracy and BLEU tasks (0-100). Source: huggingface.co. Saturation forecast: Around August 2027. 32 models tracked. Updated 22 Nov 2026.

Top models

#ModelScoreOverall rank
1Ayb-0.5-9B77.42
2Ayb-0.5-4B73.75
3GPT-6.1 Sol72.5#15
4DeepSeek V4.1 Flash71.18#69
5Ayb-0.5-0.8B69.33
6Qwen 3.5 27B69.17#380
7Gemini 2.5 Pro68.79#280
8GPT-5.4 Mini68.31#409
9GPT-5.2 Pro67.84#139
10Claude Sonnet 467.71#414
11Gemma 4 31B (IT)67.58#354
12Claude 3.7 Sonnet67.53#455
13Gemini 3 Pro (Preview)67.36#123
14Grok 4 Fast67.05#446
15GLM-4.7 Flash66.4#1349

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=armbench-llm-v2-reading · How It Works · Data refreshed daily, snapshot 2026-11-22.