Leaderboard
On-device LLM performance rankings powered by Glicko-2
Galaxy S25 Ultra
AndroidRank
#25
Rating
1,897
±13 RD
Win Rate
88.3%
Conservative Rating
1,871
TG Rating
1,942
PP Rating
1,832
Matches
1,577
Record
1392W – 185L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| FedPPO-Confused-Pythia-70M-a0.Q2_K | 477.25 | 2095.90 | 477.25 | 2095.90 | 1 |
| tinygemma3-Q8_0 | 251.81 | 13650.76 | 251.81 | 13650.76 | 1 |
| Qwen3-0.6B-Q4_0 | 106.78 | 414.66 | 107.63 | 433.67 | 2 |
| SmolVLM-500M-Instruct-Q8_0 | 97.43 | 390.77 | 97.43 | 390.77 | 1 |
| gemma-3-270m-it-F16 | 90.21 | 412.06 | 90.21 | 412.06 | 1 |
| chatgpt-5-q8_0 | 86.03 | 394.83 | 86.03 | 394.83 | 1 |
| smollm2-360m-instruct-q8_0 | 71.56 | 235.46 | 71.56 | 235.46 | 1 |
| gemma-3-1B-it-QAT-Q4_0 | 55.99 | 394.67 | 55.99 | 394.67 | 1 |
| gemma-3-1b-it.Q2_K | 53.10 | 189.82 | 53.10 | 189.82 | 1 |
| llama-3.2-1b-instruct-q4_k_m | 52.25 | 150.10 | 52.25 | 150.10 | 1 |
| SmolVLM-500M-Instruct-f16 | 48.29 | 763.18 | 52.56 | 835.92 | 2 |
| gemma-3-1b-it.Q8_0 | 47.70 | 246.76 | 47.70 | 246.76 | 1 |
| llama-3.2-1b-instruct.Q8_0 | 40.40 | 149.96 | 40.40 | 149.96 | 1 |
| llama-3.2-1b-instruct-q8_0 | 32.72 | 179.07 | 41.03 | 675.36 | 4 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 31.12 | 99.49 | 31.12 | 99.49 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 28.83 | 131.47 | 32.50 | 586.56 | 7 |
| gemma-3-1b-it.fp16 | 26.84 | 63.12 | 26.84 | 63.12 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 25.92 | 92.18 | 27.78 | 116.07 | 3 |
| Qwen3-4B.Q6_K | 20.53 | 19.20 | 31.11 | 22.70 | 2 |
| gemma-3-4b-it-Q4_0 | 20.26 | 97.61 | 22.00 | 101.84 | 2 |
| qwen2.5-3b-instruct-q3_k_m | 20.08 | 31.01 | 20.08 | 31.01 | 1 |
| gemma-3-4B-it-QAT-Q4_0 | 19.05 | 98.77 | 19.81 | 104.22 | 2 |
| Qwen3VL-2B-Thinking-Q4_K_M | 18.16 | 73.30 | 18.16 | 73.30 | 1 |
| Qwen3-4B.Q4_K_M | 17.90 | 27.96 | 17.90 | 27.96 | 1 |
| medgemma-4b-it-Q4_K_M | 16.66 | 28.59 | 16.66 | 28.59 | 1 |
| gemma-3n-E2B-it-Q8_0 | 16.20 | 55.76 | 16.20 | 55.76 | 1 |
| gemma-3n-E2B-it-Q8_0 | 16.06 | 43.55 | 16.06 | 43.55 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 15.60 | 21.35 | 18.54 | 37.68 | 16 |
| gemma-2-2b-it-Q6_K | 14.80 | 30.73 | 21.25 | 50.68 | 15 |
| gemma-3-4b-it-q4_0 | 14.67 | 88.41 | 15.42 | 99.42 | 2 |
| qwen2.5-3b-instruct-q5_k_m | 14.57 | 25.58 | 20.30 | 83.00 | 17 |
| llama-3.2-3b-instruct-q4_k_m | 14.55 | 38.99 | 14.55 | 38.99 | 1 |
| Phi-4-mini-instruct-abliterated-Q4_K_M | 14.53 | 21.46 | 14.53 | 21.46 | 1 |
| qwen2.5-coder-3b-instruct-q6_k | 14.36 | 19.90 | 14.36 | 19.90 | 1 |
| gemma-3-4b-it-Q4_K_M | 14.15 | 25.37 | 14.15 | 25.37 | 1 |
| gemma-3-4b-it.Q3_K_S | 13.68 | 17.98 | 13.68 | 17.98 | 1 |
| google_gemma-3-4b-it-Q8_0 | 13.64 | 51.80 | 13.64 | 51.80 | 1 |
| Llama-3.2-3B-Instruct-uncensored.Q5_K_M | 13.59 | 22.82 | 13.59 | 22.82 | 1 |
| Gemma-3-4b-it-abliterated-Q4_0 | 13.50 | 201.71 | 13.50 | 201.71 | 1 |
| gemma-3-4b-it.Q8_0 | 13.11 | 50.36 | 13.11 | 50.36 | 1 |
| google_gemma-3n-E2B-it-Q8_0 | 12.96 | 36.84 | 12.96 | 36.84 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 12.90 | 30.45 | 17.33 | 70.94 | 22 |
| VLSP-qwen3-4b-NHK.Q4_K_S | 12.83 | 27.34 | 12.83 | 27.34 | 1 |
| Qwen3-8B-Q4_0 | 12.49 | 53.17 | 12.49 | 53.17 | 1 |
| qwen2.5-3b-instruct-q6_k | 12.45 | 17.97 | 12.45 | 17.97 | 1 |
| Gemma-3-it-4B-Uncensored-D_AU-Q8_0 | 11.92 | 42.16 | 11.92 | 42.16 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 11.74 | 33.38 | 17.66 | 44.68 | 7 |
| gemma-3n-E4B-it-Q4_K_M | 11.39 | 25.76 | 13.52 | 37.80 | 5 |
| Gemmasutra-Small-4B-v1.i1-Q6_K | 11.38 | 18.88 | 11.38 | 18.88 | 1 |
| gemma-3n-E4B-it-Q2_K_L | 11.36 | 15.75 | 11.36 | 15.75 | 1 |
1–50 of 78 rows
1 / 2
Head-to-Head Record
1–50 of 347 rows
1 / 7
Performance by App Version
ImprovedRegressed