Leaderboard
On-device LLM performance rankings powered by Glicko-2
POCO F7 Ultra
AndroidRank
#23
Rating
1,906
±17 RD
Win Rate
89.2%
Conservative Rating
1,872
TG Rating
1,917
PP Rating
1,846
Matches
952
Record
849W – 103L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| llama-3.2-1b-instruct-q8_0 | 30.40 | 759.75 | 30.40 | 759.75 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 24.41 | 80.61 | 24.41 | 80.61 | 1 |
| Llama-3.2-3B-Instruct-uncensored-Q4_K_M | 17.71 | 29.58 | 17.71 | 29.58 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 16.75 | 23.00 | 16.75 | 23.00 | 1 |
| gemma-2-2b-it-Q6_K | 15.07 | 51.15 | 15.07 | 51.15 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 12.64 | 28.91 | 14.20 | 37.37 | 2 |
| DeepSeek-R1-0528-Qwen3-8B-Q4_0 | 9.45 | 35.49 | 9.45 | 35.49 | 1 |
| DeepSeek-R1-0528-Qwen3-8B-Q6_K | 4.39 | 10.50 | 4.39 | 10.50 | 1 |
| gemma-3-12b-it-Q4_K_M | 3.32 | 7.54 | 3.32 | 7.54 | 1 |
Head-to-Head Record
1–50 of 246 rows
1 / 5
Performance by App Version
ImprovedRegressed