Leaderboard
On-device LLM performance rankings powered by Glicko-2
Xiaomi 15
AndroidRank
#57
Rating
1,774
±18 RD
Win Rate
76.4%
Conservative Rating
1,737
TG Rating
1,788
PP Rating
1,780
Matches
789
Record
603W – 186L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| qwen2.5-0.5b-instruct-q4_k_m | 62.65 | 145.83 | 62.65 | 145.83 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 24.87 | 71.12 | 24.87 | 71.12 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 18.11 | 49.52 | 18.11 | 49.52 | 1 |
| Qwen3-4B.Q4_K_M | 15.83 | 23.58 | 15.83 | 23.58 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 14.36 | 86.53 | 14.36 | 86.53 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 14.35 | 97.76 | 14.35 | 97.76 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 12.34 | 23.17 | 12.34 | 23.17 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 10.08 | 25.69 | 13.46 | 37.85 | 2 |
| gemma-2-2b-it-Q6_K | 10.06 | 46.41 | 16.86 | 92.84 | 3 |
| Qwen3.5-0.8B-Q4_K_M | 9.57 | 136.64 | 9.57 | 136.64 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 7.73 | 16.27 | 8.67 | 23.07 | 2 |
| Phi-4-mini-instruct-Q4_K_M | 7.53 | 25.35 | 7.53 | 25.35 | 1 |
| DeepSeek-R1-Distill-Llama-8B-Q6_K | 5.52 | 6.72 | 5.52 | 6.72 | 1 |
| DeepSeek-R1-Distill-Qwen-7B-IQ2_M | 3.00 | 3.84 | 3.00 | 3.84 | 1 |
| DeepSeek-R1-Distill-Qwen-7B-Uncensored.i1-IQ1_M | 2.70 | 4.59 | 2.70 | 4.59 | 1 |
Head-to-Head Record
1–50 of 244 rows
1 / 5
Performance by App Version
ImprovedRegressed