Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 17
iOSRank
#12
Rating
1,954
±15 RD
Win Rate
93.8%
Conservative Rating
1,923
TG Rating
1,931
PP Rating
1,962
Matches
1,134
Record
1064W – 70L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| lille-130m-instruct-f32 | 92.31 | 374.71 | 92.31 | 374.71 | 1 |
| gemma-3-1b-it-Q4_0 | 55.25 | 95.18 | 58.61 | 100.83 | 2 |
| Qwen3.5-0.8B-Q4_0 | 45.89 | 757.16 | 45.89 | 757.16 | 1 |
| llama-3.2-1b-instruct-q8_0 | 45.84 | 773.37 | 46.66 | 826.28 | 3 |
| LFM2-2.6B-Exp-Q4_K_M | 35.44 | 296.23 | 35.44 | 296.23 | 1 |
| Qwen3.5-2B.Q2_K | 33.23 | 409.43 | 33.23 | 409.43 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 26.52 | 341.31 | 26.52 | 341.31 | 1 |
| gemma-2-2b-it.Q6_K | 26.14 | 345.06 | 26.14 | 345.06 | 1 |
| gemma-2-2b-it-Q6_K | 25.29 | 328.67 | 27.22 | 360.23 | 6 |
| qwen2.5-3b-instruct-q5_k_m | 24.47 | 231.52 | 24.47 | 231.52 | 1 |
| gemma-3-4b-it-IQ4_NL | 23.07 | 230.01 | 23.07 | 230.01 | 1 |
| Huihui-Qwen3-4B-Thinking-2507-abliterated.Q4_K_M | 21.70 | 187.43 | 21.70 | 187.43 | 1 |
| Qwen3-VL-4B-Instruct-Q4_K_M | 21.69 | 197.24 | 21.69 | 197.24 | 1 |
| Qwen3.5-2B-Q8_0 | 21.41 | 400.66 | 21.41 | 400.66 | 1 |
| Qwen3-4B-Instruct-2507-UD-Q4_K_XL | 21.36 | 190.17 | 22.05 | 204.64 | 2 |
| Qwen3.5-2B.Q8_0 | 21.34 | 411.90 | 21.34 | 411.90 | 1 |
| Qwen.Qwen3-VL-Embedding-2B.Q8_0 | 17.98 | 304.30 | 17.98 | 304.30 | 1 |
| gemma-3-4b-it-q4_0 | 17.78 | 210.10 | 17.78 | 210.10 | 1 |
| gemma-3-4b-it-abliterated-v2.q6_k | 17.71 | 226.47 | 17.71 | 226.47 | 1 |
| guanaco-7b-uncensored.Q2_K | 17.19 | 107.11 | 17.33 | 107.71 | 2 |
| Llama-3.2-3B-Instruct-Q6_K | 16.22 | 133.76 | 21.64 | 251.44 | 2 |
| gemma-3n-E2B-it-Q5_K_S | 16.16 | 159.10 | 16.16 | 159.10 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 13.07 | 248.66 | 13.07 | 248.66 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 12.98 | 211.14 | 12.98 | 211.14 | 1 |
| Llama-3.2-3B-Instruct-Q4_0 | 12.50 | 18.39 | 12.50 | 18.39 | 1 |
| Qwen3.5-4B-IQ4_NL | 12.14 | 142.62 | 12.14 | 142.62 | 1 |
| Qwen3.5-4B-Q4_0 | 12.01 | 153.24 | 12.17 | 153.92 | 2 |
| Crow-4B-Opus-4.6-Distill-Heretic_Qwen3.5.i1-Q4_K_M | 11.36 | 139.64 | 11.36 | 139.64 | 1 |
| Qwen3.5-4B-Q4_1 | 10.22 | 134.61 | 10.22 | 134.61 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 9.87 | 14.82 | 9.87 | 14.82 | 1 |
| Qwen2.5-VL-7B-Instruct-iq2_m | 9.47 | 76.43 | 9.47 | 76.43 | 1 |
Head-to-Head Record
1–50 of 318 rows
1 / 7
Performance by App Version
ImprovedRegressed