Leaderboard
On-device LLM performance rankings powered by Glicko-2
Galaxy S24 Ultra
AndroidRank
#44
Rating
1,808
±13 RD
Win Rate
79.7%
Conservative Rating
1,782
TG Rating
1,834
PP Rating
1,730
Matches
1,506
Record
1201W – 305L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| gemma-3-270m-it-qat-UD-Q6_K_XL | 75.96 | 260.08 | 75.96 | 260.08 | 1 |
| FastThink-0.5B-Tiny.i1-Q4_K_M | 64.45 | 174.78 | 64.45 | 174.78 | 1 |
| FastThink-0.5B-Tiny.i1-Q4_1 | 61.15 | 223.79 | 66.67 | 250.59 | 7 |
| Qwen3-0.6B.Q6_K | 43.35 | 85.06 | 43.35 | 85.06 | 1 |
| Qwen3-0.6B-Q8_0 | 43.35 | 160.02 | 43.35 | 160.02 | 1 |
| gemma-3-1b-it-Q8_0 | 31.95 | 202.46 | 31.95 | 202.46 | 1 |
| gemma-3-1b-it.Q8_0 | 28.96 | 186.70 | 28.96 | 186.70 | 1 |
| PyThagorean-Tiny.i1-IQ1_S | 28.45 | 55.37 | 28.45 | 55.37 | 1 |
| SmolVLM2-256M-Video-Instruct-f16 | 28.12 | 144.15 | 28.12 | 144.15 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-IQ4_NL | 26.67 | 109.51 | 26.67 | 109.51 | 1 |
| Janus-Pro-1B-LM.i1-Q4_1 | 25.99 | 78.48 | 28.75 | 81.13 | 3 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q2_K | 24.83 | 43.57 | 24.83 | 43.57 | 1 |
| gemma-3-1b-it.Q4_K_S | 23.14 | 79.59 | 23.14 | 79.59 | 1 |
| Bielik-1.5B-v3.0-Instruct.Q8_0 | 23.07 | 89.61 | 23.07 | 89.61 | 1 |
| Qwen3-1.7B-Q3_K_M | 22.34 | 40.89 | 22.34 | 40.89 | 1 |
| Llama-3.2-1B-Instruct.IQ1_M | 22.18 | 34.42 | 22.18 | 34.42 | 1 |
| Janus-Pro-1B-LM.i1-IQ1_S | 21.89 | 41.65 | 21.89 | 41.65 | 1 |
| deepscaler-1.5b-preview-q8_0 | 21.13 | 81.43 | 21.13 | 81.43 | 1 |
| SmolVLM2-2.2B-Instruct-Q4_K_M | 20.89 | 56.66 | 20.89 | 56.66 | 1 |
| Qwen3-1.7B-Q6_K | 20.73 | 34.57 | 20.73 | 34.57 | 1 |
| LFM2-2.6B-Q4_0 | 20.15 | 125.56 | 20.15 | 125.56 | 1 |
| llama-3.2-1b-instruct-q8_0 | 19.27 | 108.05 | 81.53 | 129.44 | 7 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 19.19 | 73.07 | 21.94 | 84.68 | 2 |
| gemma-3-1b-it-q8_0 | 19.08 | 107.92 | 19.08 | 107.92 | 1 |
| ReasoningCore-Llama-3.2-1B-RE01-0.i1-Q6_K | 18.99 | 47.62 | 18.99 | 47.62 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 18.27 | 82.00 | 18.27 | 82.00 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 17.64 | 66.50 | 18.86 | 74.18 | 4 |
| qwen2.5-1.5b-instruct-q8_0 | 16.61 | 79.85 | 25.63 | 108.41 | 7 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_S | 16.46 | 29.06 | 16.46 | 29.06 | 1 |
| LFM2-VL-1.6B-Q8_0 | 15.75 | 112.67 | 15.75 | 112.67 | 1 |
| Qwen3-1.7B-Q4_0 | 15.45 | 258.37 | 15.45 | 258.37 | 1 |
| Llama-3.2-3B-Instruct-uncensored-Q4_0 | 15.34 | 58.34 | 15.34 | 58.34 | 1 |
| Vikhr-Gemma-2B-instruct-Q4_1 | 15.32 | 46.24 | 16.39 | 50.12 | 2 |
| Llama-3.2-3B-Instruct-uncensored-IQ4_XS | 14.65 | 21.63 | 14.65 | 21.63 | 1 |
| SmolVLM2-2.2B-Instruct-Q8_0 | 14.22 | 51.46 | 14.45 | 69.53 | 2 |
| qwen2.5-coder-3b-instruct-q4_k_m | 12.35 | 25.44 | 12.35 | 25.44 | 1 |
| Qwen3-4B-Q4_K_M | 11.75 | 18.22 | 11.75 | 18.22 | 1 |
| Phi-3-Context-Obedient-RAG.Q5_K_S | 11.27 | 15.66 | 11.27 | 15.66 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 11.12 | 28.94 | 13.50 | 39.28 | 4 |
| EXAONE-Deep-2.4B-Q8_0 | 10.58 | 40.62 | 10.61 | 42.88 | 2 |
| qwen2.5-3b-instruct-q5_k_m | 10.50 | 19.20 | 14.48 | 27.25 | 11 |
| gemma-2-2b-it-Q6_K | 10.38 | 26.40 | 15.07 | 65.20 | 25 |
| gemma-3-4b-it-qat-UD-Q4_K_XL | 10.22 | 21.19 | 10.22 | 21.19 | 1 |
| gemma-3-4b-it.Q4_K_S | 10.04 | 23.32 | 10.04 | 23.32 | 1 |
| gemma-3-4b-it-qat-Q4_K_M | 9.96 | 23.18 | 9.96 | 23.18 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 9.76 | 17.56 | 14.54 | 22.52 | 17 |
| Phi-3.5-mini-instruct.Q8_0 | 9.30 | 35.48 | 9.30 | 35.48 | 1 |
| Qwen3-4B-Instruct-2507-Q4_K_M | 9.11 | 24.58 | 10.39 | 28.33 | 2 |
| Llama-3.2-3B-Instruct-Q6_K | 9.01 | 16.54 | 13.04 | 31.51 | 22 |
| SmolLM2-1.7B-Instruct-IQ3_M | 8.94 | 12.34 | 8.94 | 12.34 | 1 |
1–50 of 94 rows
1 / 2
Head-to-Head Record
1–50 of 339 rows
1 / 7
Performance by App Version
ImprovedRegressed