Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 14
iOSRank
#29
Rating
1,901
±19 RD
Win Rate
88.8%
Conservative Rating
1,863
TG Rating
1,825
PP Rating
1,918
Matches
729
Record
647W – 82L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 2361.59 | 336949.15 | 2391.39 | 367971.98 | 4 |
| Qwen3-0.6B-Q4_K_M | 50.31 | 81.66 | 50.31 | 81.66 | 1 |
| Qwen3-0.6B.Q4_K_M | 49.86 | 891.35 | 49.86 | 891.35 | 1 |
| Qwen3-0.6B-Q4_0 | 49.32 | 80.82 | 49.32 | 80.82 | 1 |
| Qwen3-0.6B.Q6_K | 48.56 | 917.26 | 48.56 | 917.26 | 1 |
| qwen2.5-0.5b-instruct-q8_0 | 44.65 | 1154.22 | 44.65 | 1154.22 | 1 |
| gemma-3-270m-it-F16 | 29.51 | 1622.65 | 29.51 | 1622.65 | 1 |
| gemma-3-1b-it.Q2_K | 26.44 | 43.85 | 27.57 | 45.55 | 2 |
| gemma-3-1b-it.Q4_K_M | 24.56 | 339.60 | 30.14 | 646.49 | 2 |
| google_gemma-3-1b-it-Q8_0 | 24.51 | 721.65 | 24.51 | 721.65 | 1 |
| gemma-3-1b-it.Q8_0 | 24.50 | 671.14 | 24.66 | 716.51 | 4 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M | 21.15 | 29.93 | 21.15 | 29.93 | 1 |
| Qwen3.5-0.8B_Abliterated.IQ4_XS | 20.80 | 433.93 | 20.80 | 433.93 | 1 |
| llama-3.2-1b-instruct-q8_0 | 20.46 | 501.96 | 20.46 | 501.96 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-UD-Q4_K_XL | 19.22 | 28.38 | 19.22 | 28.38 | 1 |
| Qwen3-1.7B.Q4_K_M | 16.86 | 23.70 | 16.86 | 23.70 | 1 |
| gemma-2-2b-it-Q6_K | 13.02 | 214.35 | 13.15 | 230.41 | 3 |
| DeepSeek-R1-Distill-Qwen-1.5B-IQ2_M | 11.58 | 11.20 | 11.58 | 11.20 | 1 |
| gemma-3-4b-it.Q2_K | 10.74 | 120.95 | 10.74 | 120.95 | 1 |
| gemma-3-4b-it-Q4_K_M | 10.26 | 157.67 | 10.26 | 157.67 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 10.19 | 147.63 | 10.69 | 155.52 | 2 |
| DeepSeek-R1-Distill-Qwen-7B-IQ2_M | 4.97 | 54.81 | 5.33 | 60.35 | 2 |
| Qwen3.5-4B-IQ4_NL | 3.56 | 53.75 | 3.56 | 53.75 | 1 |
Head-to-Head Record
1–50 of 212 rows
1 / 5
Performance by App Version
ImprovedRegressed