Leaderboard
On-device LLM performance rankings powered by Glicko-2
REDMAGIC 9 Pro
AndroidRank
#71
Rating
1,723
±14 RD
Win Rate
71.5%
Conservative Rating
1,694
TG Rating
1,662
PP Rating
1,853
Matches
1,333
Record
953W – 380L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| llama-3.2-1b-instruct-q8_0 | 23.29 | 426.44 | 23.29 | 426.44 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 19.34 | 301.64 | 19.34 | 301.64 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 16.10 | 248.52 | 16.10 | 248.52 | 1 |
| gemma-3-4b-it.Q2_K | 10.88 | 16.76 | 10.88 | 16.76 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 9.90 | 22.45 | 12.50 | 26.94 | 7 |
| qwen2.5-3b-instruct-q5_k_m | 9.73 | 19.92 | 11.51 | 45.69 | 6 |
| Llama-3.2-3B-Instruct-Q6_K | 6.20 | 30.14 | 10.03 | 32.19 | 2 |
| DeepSeek-V2-Lite.Q2_K | 5.55 | 21.23 | 5.55 | 21.23 | 1 |
| Qwen3.5-2B.Q8_0 | 5.43 | 186.30 | 5.43 | 186.30 | 1 |
| Qwen3.5-0.8B-Q4_0 | 5.36 | 251.54 | 5.36 | 251.54 | 1 |
| DeepSeek-R1-Distill-Qwen-7B-Q6_K | 5.17 | 10.18 | 5.17 | 10.18 | 1 |
| DeepSeek-R1-Distill-Llama-8B-Q2_K_L | 4.98 | 8.59 | 4.98 | 8.59 | 1 |
| Qwen3.5-0.8B-Q4_K_M | 4.96 | 48.77 | 4.96 | 48.77 | 1 |
| Qwen3.5-0.8B-Q3_K_S | 4.83 | 80.07 | 4.83 | 80.07 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 4.33 | 68.84 | 4.33 | 68.84 | 1 |
| Qwen3.5-2B.Q3_K_S | 3.57 | 38.29 | 3.57 | 38.29 | 1 |
| gemma-2-2b-it-Q6_K | 3.02 | 44.80 | 3.02 | 44.80 | 1 |
| llama-3.1-8b-instruct-q4_0 | 2.58 | 26.00 | 2.58 | 26.00 | 1 |
| Qwen3.5-4B-Q8_0 | 2.54 | 76.69 | 2.54 | 76.69 | 1 |
| Qwen3.5-4B-UD-IQ2_XXS | 2.41 | 10.85 | 2.41 | 10.85 | 1 |
| Qwen3.5-4B-UD-IQ3_XXS | 2.14 | 9.40 | 2.14 | 9.40 | 1 |
| Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING.i1-Q4_1 | 1.79 | 13.87 | 1.79 | 13.87 | 1 |
| Qwen3.5-4B-Q4_K_M | 1.78 | 7.53 | 1.78 | 7.53 | 1 |
| Qwen3.5-9B-Q3_K_S | 1.57 | 7.38 | 1.57 | 7.38 | 1 |
| Qwen3.5-9B-Q4_K_M | 1.05 | 3.95 | 1.05 | 3.95 | 1 |
Head-to-Head Record
1–50 of 318 rows
1 / 7
Performance by App Version
ImprovedRegressed