Leaderboard
On-device LLM performance rankings powered by Glicko-2
Xiaomi 15 Ultra
AndroidRank
#65
Rating
1,754
±17 RD
Win Rate
74.5%
Conservative Rating
1,720
TG Rating
1,711
PP Rating
1,771
Matches
910
Record
678W – 232L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| gemma-3-270m-it-F16 | 83.45 | 348.47 | 83.45 | 348.47 | 1 |
| RakutenAI-2.0-mini-instruct-Q4_0 | 53.49 | 187.02 | 53.49 | 187.02 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 18.77 | 53.76 | 18.77 | 53.76 | 1 |
| Qwen3-4B.Q6_K | 11.64 | 24.95 | 11.64 | 24.95 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 10.53 | 19.72 | 15.99 | 24.30 | 4 |
| Llama-3-8B-Japanese-Instruct-Q4_K_M | 9.12 | 16.21 | 9.12 | 16.21 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 8.57 | 17.15 | 9.13 | 20.92 | 3 |
| DeepSeek-R1-Distill-Qwen-7B-Q4_K_M | 8.48 | 12.08 | 8.48 | 12.08 | 1 |
| MiMo-7B-RL-Q4_K_M | 7.63 | 11.82 | 7.63 | 11.82 | 1 |
| gemma-2-2b-it-Q6_K | 6.68 | 82.41 | 6.68 | 82.41 | 1 |
| dolphin-llama3-zh-cn-uncensored-unsloth.Q4_K_M | 5.40 | 10.46 | 5.40 | 10.46 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 5.38 | 59.29 | 5.38 | 59.29 | 1 |
Head-to-Head Record
1–50 of 242 rows
1 / 5
Performance by App Version
ImprovedRegressed