Leaderboard
On-device LLM performance rankings powered by Glicko-2
Xiaomi 14 Ultra
AndroidRank
#58
Rating
1,786
±25 RD
Win Rate
77.6%
Conservative Rating
1,735
TG Rating
1,822
PP Rating
1,676
Matches
425
Record
330W – 95L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 18.88 | 79.64 | 20.66 | 86.45 | 2 |
| qwen2.5-3b-instruct-q5_k_m | 13.46 | 21.53 | 13.46 | 21.53 | 1 |
| chocolatine-3b-instruct-dpo-revised-q4_k_m | 11.38 | 22.92 | 11.38 | 22.92 | 1 |
| Huihui-Qwen3-VL-4B-Instruct-abliterated-Q4_K_M | 9.80 | 25.75 | 9.80 | 25.75 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 8.66 | 18.90 | 10.80 | 21.73 | 4 |
| Qwen3.5-2B-IQ4_NL | 4.51 | 99.28 | 4.51 | 99.28 | 1 |
| Qwen3-8B.Q4_K_M | 3.54 | 11.23 | 3.54 | 11.23 | 1 |
| Dark-Desires-12B-v1.0-Q2_K | 2.54 | 3.93 | 2.54 | 3.93 | 1 |
| Qwen3.5-4B-IQ4_NL | 1.89 | 38.49 | 1.89 | 38.49 | 1 |
Head-to-Head Record
1–50 of 113 rows
1 / 3
Performance by App Version
ImprovedRegressed