ArmBench-LLM - Generation: leaderboard
ArmBench-LLM category: Armenian text generation, the mean BLEU against reference texts on email summarization, conversation summarization, simple QA and paraphrase tasks. 0-100 BLEU, higher is better; every listed model scores below 15.
Metric: BLEU (0-100). Source: huggingface.co. Saturation forecast: Not forecast. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 14.37 | #255 |
| 2 | Claude 3.7 Sonnet | 12.89 | #259 |
| 3 | GPT-5.4 Mini | 12.13 | #228 |
| 4 | Claude Sonnet 4 | 11.8 | #207 |
| 5 | GPT-5.4 Nano | 11.52 | #359 |
| 6 | Grok 4.1 Fast | 11.32 | #227 |
| 7 | Gemma 3 12B (IT) | 11.31 | #672 |
| 8 | GPT-5.2 Pro | 11.08 | #70 |
| 9 | Gemma 4 31B (IT) | 10.87 | #178 |
| 10 | Grok 4 Fast | 10.72 | #266 |
| 11 | Gemma 3 27B (IT) | 10.19 | #528 |
| 12 | GLM-5 | 9.93 | #149 |
| 13 | Mistral Large 3 | 9.7 | #410 |
| 14 | GLM-4.7 Flash | 9.52 | #543 |
| 15 | Gemini 3 Pro (Preview) | 9.42 | #71 |
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-generation · How It Works · Data refreshed daily, snapshot 2026-11-05.