Edge LLM Leaderboard: Raspberry Pi 5 - Prefill: leaderboard
Edge LLM Raspberry Pi 5 leaderboard measuring prompt prefill throughput in tokens per second for quantized local models.
Metric: Prefill (tokens/s). Source: huggingface.co. 128 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gpt2 (Q4_0_4_4, llama_cpp) | 582.32 | |
| 2 | SmolLM2-135M-Instruct (Q4_0_4_4, llama_cpp) | 381.67 | |
| 3 | gpt2 (Q8_0, llama_cpp) | 302.93 | |
| 4 | gpt2 (Q4_K_M, llama_cpp) | 269.91 | |
| 5 | SmolLM2-135M-Instruct (Q8_0, llama_cpp) | 212.15 | |
| 6 | Qwen2.5-0.5B (Q4_0_4_4, llama_cpp) | 189.87 | |
| 7 | gpt2-medium (Q4_0_4_4, llama_cpp) | 178.73 | |
| 8 | SmolLM2-360M-Instruct (Q4_0_4_4, llama_cpp) | 173.28 | |
| 9 | SmolLM2-135M-Instruct (Q4_K_M, llama_cpp) | 153.44 | |
| 10 | Llama-3.2-1B (Q4_0_4_4, llama_cpp) | 86.77 |
Interactive version: theaggregate.ai/benchmark?slug=edge-llm-leaderboard-raspberry-pi-5-prefill · How It Works · Data refreshed daily, snapshot 2026-10-20.