Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 16
iOSRank
#17
Rating
1,931
±14 RD
Win Rate
91.6%
Conservative Rating
1,903
TG Rating
1,924
PP Rating
1,928
Matches
1,391
Record
1274W – 117L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| google_gemma-3n-E2B-it-Q8_0 | 181.18 | 2897.66 | 181.18 | 2897.66 | 1 |
| gemma-3n-E2B-it-Q8_0 | 166.35 | 3208.17 | 166.35 | 3208.17 | 1 |
| Qwen3-0.6B-Q4_K_M | 96.29 | 1112.70 | 96.29 | 1112.70 | 1 |
| chatgpt-5-q8_0 | 73.58 | 177.50 | 73.58 | 177.50 | 1 |
| LFM2.5-1.2B-Instruct-Q4_K_M | 65.31 | 662.69 | 65.31 | 662.69 | 1 |
| DeepSeek-R1-Distill-Llama-8B-Q4_K_M | 63.50 | 528.47 | 63.86 | 623.27 | 3 |
| Qwen3-0.6B-Q8_0 | 58.84 | 1083.58 | 58.84 | 1083.58 | 1 |
| gemma-3-1B-it-QAT-Q4_0 | 58.05 | 841.01 | 58.05 | 841.01 | 1 |
| LFM2.5-1.2B-Thinking-Q5_K_M | 52.36 | 535.52 | 52.93 | 552.70 | 2 |
| Qwen2-VL-2B-Instruct-Q4_K_L | 37.95 | 420.40 | 37.95 | 420.40 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M | 32.16 | 240.09 | 38.41 | 438.73 | 2 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_M | 31.65 | 395.45 | 31.65 | 395.45 | 1 |
| qwen2.5-1.5b-thinking-q8_0 | 30.78 | 470.92 | 30.78 | 470.92 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 29.75 | 473.19 | 30.90 | 514.21 | 3 |
| qwen2.5-1.5b-instruct-q4_k_m | 29.28 | 44.85 | 29.28 | 44.85 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 28.07 | 417.97 | 28.07 | 417.97 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-uncensored.Q8_0 | 28.03 | 464.94 | 28.03 | 464.94 | 1 |
| llm-jp-3.1-1.8b-instruct4-Q8_0 | 27.45 | 431.57 | 27.45 | 431.57 | 1 |
| Qwen3.5-0.8B-Q8_0 | 25.52 | 582.91 | 25.52 | 582.91 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q2_K | 25.47 | 33.06 | 25.47 | 33.06 | 1 |
| gemma-3n-E2B-it-Q4_K_M | 24.06 | 223.58 | 24.06 | 223.58 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q2_K | 23.07 | 36.07 | 23.07 | 36.07 | 1 |
| Llama-3.2-3B-Instruct.Q4_K_M | 22.37 | 214.17 | 22.37 | 214.17 | 1 |
| Llama-3.2-3B-Instruct-Q4_K_L | 22.24 | 195.76 | 22.24 | 195.76 | 1 |
| Qwen3-1.7B.Q4_K_M | 22.13 | 36.94 | 22.13 | 36.94 | 1 |
| LFM2.5-1.2B-Instruct-BF16 | 21.98 | 673.07 | 21.98 | 673.07 | 1 |
| llama-3.2-1b-instruct-q8_0 | 21.96 | 264.34 | 21.96 | 264.34 | 1 |
| gemma-2-2b-it-Q6_K | 20.95 | 273.39 | 21.55 | 288.60 | 5 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_M | 20.04 | 30.70 | 20.04 | 30.70 | 1 |
| Llama-3.2-1B-Instruct-BF16 | 20.04 | 712.05 | 20.04 | 712.05 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 18.32 | 185.17 | 19.70 | 188.46 | 2 |
| Phi-3.5-mini-instruct.Q4_K_M | 18.32 | 155.72 | 18.32 | 155.72 | 1 |
| gemma-3-4b-it-IQ4_NL | 17.95 | 188.31 | 17.95 | 188.31 | 1 |
| gemma-3-4B-it-QAT-Q4_0 | 17.86 | 176.60 | 17.86 | 176.60 | 1 |
| Qwen3.5-2B-Q8_0 | 16.94 | 340.31 | 16.94 | 340.31 | 1 |
| Llama-3.2-3B-Instruct-Q6_K | 14.89 | 169.09 | 18.53 | 214.42 | 4 |
| Gemmasutra-Mini-2B-v1-Q6_K | 14.69 | 137.64 | 17.16 | 253.94 | 2 |
| Qwen_Qwen3-4B-Instruct-2507-Q5_K_L | 13.80 | 150.41 | 13.80 | 150.41 | 1 |
| DeepSeek-R1-ReDistill-Qwen-1.5B-v1.0-IQ3_XS | 13.60 | 13.68 | 14.08 | 14.96 | 2 |
| gemma-3n-E4B-it-Q4_K_M | 12.52 | 112.86 | 12.52 | 112.86 | 1 |
| Qwen3-4B-IQ4_NL | 12.17 | 102.79 | 12.17 | 102.79 | 1 |
| gemma-3-4b-it-Q5_K_M | 8.73 | 13.95 | 8.73 | 13.95 | 1 |
| Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M | 8.73 | 113.85 | 9.12 | 115.97 | 2 |
| Qwen3.5-4B-IQ4_NL | 8.58 | 115.16 | 9.41 | 122.50 | 2 |
| DeepSeek-R1-Distill-Qwen-7B-IQ2_M | 8.40 | 66.07 | 11.21 | 89.33 | 2 |
| gemma-3-4b-it-q4_0_s | 8.19 | 16.12 | 8.19 | 16.12 | 1 |
| gemma-3-4b-it-Q8_0 | 8.16 | 17.21 | 8.16 | 17.21 | 1 |
| Qwen3-4B.Q6_K | 7.50 | 11.20 | 7.50 | 11.20 | 1 |
| DeepSeek-R1-Distill-Qwen-7B-Q4_K_S | 7.13 | 58.90 | 7.13 | 58.90 | 1 |
| Qwen3-1.7B.fp16 | 6.45 | 19.67 | 6.45 | 19.67 | 1 |
1–50 of 53 rows
1 / 2
Head-to-Head Record
1–50 of 330 rows
1 / 7
Performance by App Version
ImprovedRegressed