SLM-RAG Arena: leaderboard
SLM-RAG Arena — pairwise-vote arena comparing small language models in retrieval-augmented generation settings.
Metric: Elo Rating. Source: huggingface.co. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | EXAONE-3.5-2.4B-Instruct | 1567.2 | #1391 |
| 2 | Phi-4 Mini Instruct | 1533.6 | #1245 |
| 3 | Qwen 2.5 3B Instruct | 1502.4 | #1162 |
| 4 | Gemma 3 4B (IT) | 1498.3 | #992 |
| 5 | SmolLM2-1.7B-Instruct | 1493.2 | #1562 |
| 6 | Gemma 2 2B (IT) | 1491.9 | #1341 |
| 7 | Qwen 3 4B | 1483.2 | #829 |
| 8 | Llama 3.2 1B Instruct | 1469.4 | #1539 |
| 9 | Qwen 3 0.6B | 1455.9 | #1459 |
| 10 | Qwen 2.5 1.5B Instruct | 1455.1 | #1407 |
| 11 | Qwen 3 1.7B | 1454.8 | #1196 |
| 12 | Llama 3.2 3B Instruct | 1435.8 | #1352 |
| 13 | Gemma 3 1B (IT) | 1421.5 | #1504 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=slm-rag-arena · How It Works · Data refreshed daily, snapshot 2026-10-20.