Leaderboard
On-device LLM performance rankings powered by Glicko-2
Galaxy S25
AndroidRank
#26
Rating
1,896
±15 RD
Win Rate
88.2%
Conservative Rating
1,867
TG Rating
1,935
PP Rating
1,807
Matches
1,276
Record
1125W – 151L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| tinystories-lay1-hs128-hd2-262K.Q4_0 | 1564.34 | 116475.35 | 1564.34 | 116475.35 | 1 |
| SmolLM2-135M-Instruct-Q4_0 | 352.05 | 1245.12 | 352.05 | 1245.12 | 1 |
| SmolLM2-135M-Instruct-Q8_0 | 244.59 | 954.16 | 244.59 | 954.16 | 1 |
| google_gemma-3-270m-it-Q4_0 | 79.21 | 1488.39 | 79.21 | 1488.39 | 1 |
| gemma-3-1b-it.Q3_K_M | 64.58 | 242.88 | 64.58 | 242.88 | 1 |
| gemma-3-1b-it-q4_0 | 50.63 | 368.92 | 50.63 | 368.92 | 1 |
| gemma-3-1b-it.Q8_0 | 46.48 | 246.54 | 47.64 | 247.51 | 2 |
| llama-3.2-1b-instruct-q8_0 | 38.53 | 127.02 | 41.54 | 176.30 | 4 |
| SmolLM2-1.7B-Instruct-Q2_K | 36.35 | 51.64 | 36.35 | 51.64 | 1 |
| LFM2.5-1.2B-Instruct-Q8_0 | 34.17 | 153.03 | 34.17 | 153.03 | 1 |
| LFM2.5-1.2B-Thinking-Q8_0 | 33.92 | 177.89 | 33.92 | 177.89 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 33.17 | 137.76 | 33.17 | 137.76 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 26.52 | 125.96 | 26.52 | 125.96 | 1 |
| Qwen_Qwen3-1.7B-Q4_0 | 22.82 | 441.92 | 26.00 | 592.17 | 2 |
| LFM2.5-1.2B-Instruct-Q5_K_M | 19.88 | 52.49 | 19.88 | 52.49 | 1 |
| LFM2-2.6B-Exp-Q8_0 | 19.79 | 54.66 | 19.80 | 55.07 | 2 |
| Phi-4-mini-instruct.Q4_K_M | 17.98 | 29.55 | 17.98 | 29.55 | 1 |
| gemma-3-4B-it-QAT-Q4_0 | 17.48 | 79.17 | 17.48 | 79.17 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-uncensored.f16 | 15.95 | 24.11 | 15.95 | 24.11 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 15.45 | 23.53 | 17.57 | 26.75 | 4 |
| Phi-3.5-mini-instruct.Q4_K_M | 14.44 | 22.72 | 18.24 | 24.80 | 3 |
| gemma-2-2b-it-Q6_K | 13.82 | 35.74 | 18.45 | 95.56 | 6 |
| gemma-3-4b-it-Q8_0 | 12.72 | 41.80 | 12.72 | 41.80 | 1 |
| Llama-3.2-1B-Instruct-IQ3_M | 12.46 | 30.00 | 12.46 | 30.00 | 1 |
| Llama-3.2-3B-Instruct-Q3_K_XL | 11.99 | 15.88 | 11.99 | 15.88 | 1 |
| Jan-nano-128k-Q2_K_L | 11.98 | 14.92 | 11.98 | 14.92 | 1 |
| Qwen2.5-THREADRIPPER-Small-AnniversaryEdition.IQ4_XS | 10.82 | 13.84 | 10.82 | 13.84 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 10.10 | 22.77 | 17.86 | 38.36 | 10 |
| gemma-3-4b-it.Q4_K_M | 9.88 | 27.49 | 9.88 | 27.49 | 1 |
| granite-3.2-8b-instruct-Q4_0 | 9.55 | 25.44 | 9.55 | 25.44 | 1 |
| Llama-3.1-Nemotron-Nano-8B-v1.i1-Q4_K_M | 8.48 | 11.15 | 8.48 | 11.15 | 1 |
| Qwen3-8B-Q4_0 | 8.45 | 78.28 | 8.45 | 78.28 | 1 |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored.Q4_K_M | 7.78 | 11.38 | 7.78 | 11.38 | 1 |
| gemma-3n-E4B-it-Q4_K_M | 6.73 | 12.83 | 6.73 | 12.83 | 1 |
| Qwen2.5-7B-Instruct-1M-Q8_0 | 6.16 | 13.14 | 6.16 | 13.14 | 1 |
| gemma-3n-E4B-it-UD-IQ2_XXS | 5.70 | 8.20 | 5.70 | 8.20 | 1 |
| Llama-3-Groq-8B-Tool-Use.Q3_K_S | 5.07 | 5.51 | 5.07 | 5.51 | 1 |
| Qwen3.5-2B-UD-Q6_K_XL | 4.46 | 68.33 | 4.46 | 68.33 | 1 |
| llama-3-Korean-Bllossom-8B.Q5_K_M | 3.86 | 6.31 | 3.86 | 6.31 | 1 |
| Qwen3.5-2B-Q4_K_M | 3.46 | 38.59 | 3.46 | 38.59 | 1 |
Head-to-Head Record
1–50 of 303 rows
1 / 7
Performance by App Version
ImprovedRegressed