Leaderboard
On-device LLM performance rankings powered by Glicko-2
POCO F5
AndroidRank
#140
Rating
1,490
±17 RD
Win Rate
49.0%
Conservative Rating
1,456
TG Rating
1,499
PP Rating
1,462
Matches
918
Record
450W – 468L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| Qwen2.5-0.5B-Instruct-Q4_K_M | 20.62 | 83.19 | 20.62 | 83.19 | 1 |
| Qwen3-0.6B-Q8_0 | 20.00 | 98.49 | 20.00 | 98.49 | 1 |
| qwen2.5-coder-1.5b-instruct-q4_k_m | 15.00 | 39.39 | 15.00 | 39.39 | 1 |
| gemma-3-1b-it-q4_0 | 13.50 | 82.53 | 13.89 | 83.77 | 2 |
| qwen2.5-coder-1.5b-instruct-q5_k_m | 13.26 | 31.65 | 13.26 | 31.65 | 1 |
| Dolphin3.0-Llama3.2-1B-Q8_0 | 12.87 | 44.91 | 12.87 | 44.91 | 1 |
| qwen2.5-coder-1.5b-instruct-q8_0 | 12.31 | 50.71 | 12.31 | 50.71 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 12.14 | 7.43 | 22.49 | 10.82 | 2 |
| Qwen3-0.6B.fp16 | 11.55 | 25.35 | 11.55 | 25.35 | 1 |
| Qwen3-1.7B-Q8_0 | 11.47 | 48.25 | 11.47 | 48.25 | 1 |
| granite-3.3-2b-instruct-Q4_K_M | 10.84 | 21.28 | 10.84 | 21.28 | 1 |
| Phi-4-mini-reasoning-Q4_K_M | 8.69 | 22.45 | 8.69 | 22.45 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 8.44 | 13.53 | 8.84 | 14.22 | 3 |
| granite-3.3-2b-instruct-Q6_K | 8.17 | 13.71 | 8.17 | 13.71 | 1 |
| Phi-4-mini-instruct.Q5_K_M | 7.81 | 16.02 | 7.81 | 16.02 | 1 |
| Qwen3-4B-Q4_K_M | 7.13 | 13.15 | 7.13 | 13.15 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 7.11 | 15.28 | 7.11 | 15.28 | 1 |
| gemma-2-2b-it-Q6_K | 6.44 | 15.18 | 7.39 | 15.25 | 3 |
| qwen2.5-3b-instruct-q5_k_m | 5.97 | 21.62 | 6.76 | 22.19 | 3 |
| Qwen3.5-2B-Q6_K | 5.67 | 38.79 | 5.67 | 38.79 | 1 |
| Qwen3-8B-Q4_K_M | 4.87 | 12.08 | 4.87 | 12.08 | 1 |
| ggml-model-Q2_K | 3.31 | 4.88 | 3.31 | 4.88 | 1 |
| Qwen3.5-4B-Q4_K_M | 3.08 | 19.26 | 3.08 | 19.26 | 1 |
Head-to-Head Record
1–50 of 246 rows
1 / 5
Performance by App Version
ImprovedRegressed