Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 16 Pro
iOSRank
#10
Rating
1,966
±15 RD
Win Rate
95.0%
Conservative Rating
1,937
TG Rating
1,961
PP Rating
1,962
Matches
1,276
Record
1212W – 64L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| Phi-4-mini-instruct-Q4_K_M | 387.42 | 6180.89 | 760.65 | 12205.03 | 2 |
| SmolLM2-135M-Instruct-Q4_0 | 193.34 | 4223.15 | 197.08 | 4262.08 | 2 |
| gemma-3n-E4B-it-IQ4_NL | 171.83 | 2378.79 | 171.83 | 2378.79 | 1 |
| SmolLM2-135M-Instruct-Q8_0 | 164.92 | 4162.77 | 169.00 | 4301.74 | 2 |
| google_gemma-3-270m-it-Q8_0 | 130.92 | 4139.69 | 130.92 | 4139.69 | 1 |
| qwen-memento-ckpt-2500-q8_0 | 120.76 | 5690.72 | 120.76 | 5690.72 | 1 |
| google_functiongemma-270m-it-Q8_0 | 119.40 | 5773.80 | 119.40 | 5773.80 | 1 |
| gemma-3-270m-it-F16 | 83.45 | 340.67 | 83.45 | 340.67 | 1 |
| google.gemma-3-270m-it.f16 | 76.61 | 3845.20 | 76.61 | 3845.20 | 1 |
| DeepSeek-R1-Distill-Llama-8B-Q4_K_M | 68.60 | 723.51 | 68.60 | 723.51 | 1 |
| Qwen3-0.6B-Q8_0 | 66.03 | 1459.38 | 66.03 | 1459.38 | 1 |
| gemma-3-1B-it-QAT-Q4_0 | 63.02 | 999.42 | 63.02 | 999.42 | 1 |
| Qwen_Qwen3-0.6B-Q8_0 | 59.85 | 1446.06 | 60.03 | 1492.10 | 4 |
| qwen-memento-ckpt-7000-q8_0 | 59.57 | 1481.66 | 59.73 | 1482.75 | 3 |
| Qwen2.5-1.5B-Instruct.Q4_K_M | 42.74 | 501.84 | 42.74 | 501.84 | 1 |
| gemma-3-1b-it.Q2_K | 39.81 | 67.00 | 39.90 | 67.56 | 2 |
| qwen-memento-ckpt-3000-q4_k_s | 39.09 | 486.24 | 39.42 | 497.88 | 3 |
| qwen-memento-ckpt-5000-q4_k_s | 38.89 | 480.26 | 39.17 | 484.52 | 2 |
| Qwen3.5-0.8B-Q4_0 | 37.73 | 599.64 | 37.73 | 599.64 | 1 |
| qwen-memento-ckpt-2000-q4_k_s | 36.61 | 492.23 | 36.61 | 492.23 | 1 |
| deepseek-r1-distill-qwen-1.5b-q6_k | 36.37 | 555.31 | 36.37 | 555.31 | 1 |
| Qwen3.5-0.8B-IQ4_NL | 36.25 | 637.28 | 36.25 | 637.28 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_L | 35.98 | 269.71 | 41.65 | 492.23 | 2 |
| Qwen_Qwen3-1.7B-Q4_K_S | 35.67 | 331.85 | 39.84 | 521.92 | 5 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q6_K | 35.40 | 459.86 | 35.40 | 459.86 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q6_K | 35.11 | 463.87 | 35.11 | 463.87 | 1 |
| SmolLM2-1.7B-Instruct-Q5_K_L | 35.04 | 374.41 | 35.04 | 374.41 | 1 |
| gemma-3-1b-it.Q8_0 | 34.93 | 98.13 | 34.93 | 98.13 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M | 34.44 | 261.79 | 42.18 | 522.36 | 4 |
| Llama-3.2-1B-Instruct.Q8_0 | 33.80 | 70.07 | 33.80 | 70.07 | 1 |
| llama-3.2-1b-instruct-q8_0 | 33.77 | 73.13 | 35.88 | 714.10 | 5 |
| DeepSeek-R1-Distill-Qwen-1.5B-uncensored.Q4_K_S | 32.63 | 49.72 | 32.63 | 49.72 | 1 |
| LFM2-VL-1.6B-Q8_0 | 30.96 | 70.41 | 30.96 | 70.41 | 1 |
| gemma-3-1b-it-Q8_0 | 30.52 | 85.81 | 30.52 | 85.81 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M | 29.99 | 46.03 | 29.99 | 46.03 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Abliterated-dpo.IQ4_XS | 29.08 | 35.41 | 29.08 | 35.41 | 1 |
| LFM2-2.6B-Exp-Q4_K_M | 27.78 | 276.33 | 27.78 | 276.33 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 27.06 | 301.27 | 28.30 | 549.74 | 4 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q2_K | 26.38 | 37.36 | 26.38 | 37.36 | 1 |
| deepseek-r1-distill-qwen-1.5b-uncensored-q8_0 | 25.41 | 55.48 | 25.41 | 55.48 | 1 |
| Bielik-1.5B-v3.0-Instruct-Q5_0 | 24.55 | 31.25 | 24.55 | 31.25 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-uncensored.Q4_K_M | 24.54 | 40.92 | 24.54 | 40.92 | 1 |
| Qwen3-1.7B-Q4_K_M | 23.92 | 39.15 | 23.92 | 39.15 | 1 |
| Llama-3.2-3B-Instruct-uncensored.IQ3_M | 23.73 | 220.65 | 23.73 | 220.65 | 1 |
| EXAONE-3.5-2.4B-Instruct-Q6_K | 23.68 | 198.72 | 23.68 | 198.72 | 1 |
| Qwen3-1.7B-Q8_0 | 23.65 | 49.01 | 23.65 | 49.01 | 1 |
| Qwen.Qwen3.5-2B.Q4_K_M | 23.47 | 369.82 | 23.47 | 369.82 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q8_0 | 23.20 | 50.20 | 23.61 | 50.81 | 2 |
| HY-MT1.5-1.8B-Q8_0 | 22.79 | 368.58 | 22.79 | 368.58 | 1 |
| Qwen3-1.7B.Q4_K_M | 22.55 | 41.50 | 22.55 | 41.50 | 1 |
1–50 of 104 rows
1 / 3
Head-to-Head Record
1–50 of 339 rows
1 / 7
Performance by App Version
ImprovedRegressed