Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 17 Pro
iOSRank
#5
Rating
1,992
±15 RD
Win Rate
97.5%
Conservative Rating
1,961
TG Rating
1,990
PP Rating
1,990
Matches
1,143
Record
1114W – 29L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| Vintern-1B-v3_5-Q8_0 | 108.01 | 2155.35 | 108.01 | 2155.35 | 1 |
| chatgpt-5-q8_0 | 102.80 | 2037.25 | 117.12 | 2154.03 | 2 |
| MinerU2.5-2509-1.2B.IQ4_XS | 93.72 | 1662.14 | 93.72 | 1662.14 | 1 |
| LFM2.5-1.2B-Thinking-Q4_0 | 86.33 | 906.92 | 86.33 | 906.92 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q2_K | 80.64 | 630.67 | 80.64 | 630.67 | 1 |
| Qwen3.5-0.8B-IQ4_NL | 53.40 | 925.02 | 53.40 | 925.02 | 1 |
| Nix2.6.Q6_K | 47.13 | 612.10 | 47.95 | 615.66 | 2 |
| qwen2.5-1.5b-instruct-q8_0 | 38.56 | 644.53 | 39.23 | 673.95 | 2 |
| granite-4.0-h-tiny-Q4_0 | 37.44 | 300.50 | 37.44 | 300.50 | 1 |
| Qwen3-0.6B.Q6_K | 33.92 | 907.69 | 33.92 | 907.69 | 1 |
| Beck-1.7B.Q8_0 | 32.98 | 602.03 | 32.98 | 602.03 | 1 |
| Qwen3.5-2B-Q4_K_M | 30.92 | 473.18 | 30.92 | 473.18 | 1 |
| Qwen3.5-0.8B-BF16 | 30.47 | 935.30 | 30.47 | 935.30 | 1 |
| gemma-2-2b-it.Q6_K | 28.35 | 360.37 | 28.35 | 360.37 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 27.53 | 270.35 | 28.13 | 283.37 | 5 |
| Gemmasutra-Mini-2B-v1-Q6_K | 27.32 | 405.07 | 27.48 | 405.21 | 2 |
| gemma-2-2b-it-Q6_K | 26.80 | 387.98 | 29.60 | 401.61 | 5 |
| Phi-3.5-mini-instruct.Q4_K_M | 26.16 | 219.68 | 26.16 | 219.68 | 1 |
| gemma-3n-E2B-it-Q4_K_M | 25.88 | 174.82 | 34.23 | 320.95 | 2 |
| gemma-3-4b-it-IQ4_NL | 24.18 | 265.89 | 24.55 | 265.99 | 3 |
| p-e-w_Qwen3-4B-Instruct-2507-heretic-Q4_K_M | 23.73 | 227.75 | 23.73 | 227.75 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 23.36 | 281.31 | 23.79 | 297.95 | 9 |
| LFM2-2.6B-Exp-Q8_0 | 22.96 | 361.62 | 22.96 | 361.62 | 1 |
| gemma-3n-E4B-it-IQ4_XS | 22.45 | 178.99 | 22.45 | 178.99 | 1 |
| Ministral-3-3B-Instruct-2512-Q5_K_M | 22.26 | 260.00 | 22.26 | 260.00 | 1 |
| Qwen3.5-2B-Uncensored-HauhauCS-Aggressive-Q4_K_M | 21.78 | 341.08 | 21.78 | 341.08 | 1 |
| Qwen3-0.6B.fp16 | 20.26 | 837.12 | 20.26 | 837.12 | 1 |
| Qwen3-8B.Q2_K | 18.59 | 112.73 | 18.59 | 112.73 | 1 |
| SauerkrautLM-7b-LaserChat.i1-IQ4_XS | 16.09 | 123.57 | 16.09 | 123.57 | 1 |
| gemma-3n-E4B-it-IQ4_NL | 15.70 | 162.29 | 20.42 | 162.98 | 2 |
| SauerkrautLM-7b-LaserChat.IQ4_NL | 15.58 | 127.41 | 15.58 | 127.41 | 1 |
| Qwen3.5-4B_Abliterated.Q4_K_S | 14.92 | 191.27 | 14.92 | 191.27 | 1 |
| DeepSeek-R1-Distill-Qwen-7B-Q4_K_S | 14.88 | 119.80 | 14.88 | 119.80 | 1 |
| L3-8B-Stheno-v3.3-32K-NEO-V1-D_AU-IQ4_XS-imat13 | 14.78 | 120.29 | 14.78 | 120.29 | 1 |
| Qwen3-8B-f16:Q3_K_S | 14.69 | 111.03 | 14.69 | 111.03 | 1 |
| Dolphin3.0-Qwen2.5-3b-Q6_K | 14.46 | 19.12 | 14.46 | 19.12 | 1 |
| Qwen3.5-2B-UD-Q8_K_XL | 14.26 | 367.18 | 14.26 | 367.18 | 1 |
| Llama-3.2-3B-Instruct-uncensored-Q6_K | 14.12 | 19.05 | 14.12 | 19.05 | 1 |
| Qwen3.5-4B-Q4_K_M | 13.35 | 173.79 | 13.35 | 173.79 | 1 |
| DeepSeek-R1-0528-Qwen3-8B-IQ4_NL | 12.74 | 110.35 | 12.74 | 110.35 | 1 |
| gemma-3-4b-it-UD-Q8_K_XL | 11.98 | 251.59 | 11.98 | 251.59 | 1 |
| Qwen2.5-7B-Instruct-Q5_K_M | 11.36 | 120.24 | 11.36 | 120.24 | 1 |
| Meta-Llama-3.1-8B-Instruct.Q4_K_M | 11.34 | 103.33 | 11.34 | 103.33 | 1 |
| EVA-abliterated-Qwen2.5-7B.i1-Q5_K_M | 11.28 | 115.54 | 11.28 | 115.54 | 1 |
| Qwen3-8B-Q4_K_M | 10.75 | 100.79 | 12.33 | 113.75 | 2 |
| Qwen3.5-4B-Q8_0 | 10.48 | 182.95 | 10.48 | 182.95 | 1 |
| ARM-DS-R1-Qwen3-8B-ArliAI-RpR-v4-Small-Q4_0-imat | 10.23 | 65.30 | 13.96 | 119.75 | 2 |
| Qwen3.5-4B-IQ4_NL | 10.14 | 134.72 | 14.36 | 182.66 | 2 |
| Qwen3.5-4B-Q5_K_M | 9.93 | 147.20 | 10.43 | 152.39 | 2 |
| DeepSeek-R1-0528-Qwen3-8B-IQ4_XS | 9.65 | 56.86 | 13.65 | 105.56 | 2 |
1–50 of 52 rows
1 / 2
Head-to-Head Record
1–50 of 282 rows
1 / 6
Performance by App Version
ImprovedRegressed