Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 12
iOSRank
#60
Rating
1,787
±28 RD
Win Rate
77.8%
Conservative Rating
1,731
TG Rating
1,743
PP Rating
1,893
Matches
351
Record
273W – 78L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| SmolVLM-256M-Instruct-Q8_0 | 125.99 | 276.00 | 125.99 | 276.00 | 1 |
| gemma-3-270m-it-UD-Q8_K_XL | 60.34 | 269.96 | 60.34 | 269.96 | 1 |
| Qwen3-0.6B-Q4_K_M | 38.20 | 62.87 | 38.20 | 62.87 | 1 |
| Qwen_Qwen3-0.6B-Q8_0 | 36.59 | 476.73 | 36.59 | 476.73 | 1 |
| gemma-3-1b-it.Q5_K_M | 28.39 | 329.46 | 28.39 | 329.46 | 1 |
| gemma-3-1B-it-QAT-Q4_0 | 28.21 | 251.03 | 28.21 | 251.03 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-Q4_0 | 25.75 | 194.56 | 25.75 | 194.56 | 1 |
| google_gemma-3-1b-it-qat-Q6_K | 23.19 | 44.50 | 23.19 | 44.50 | 1 |
| gemma-3-1b-it.Q2_K | 22.78 | 37.95 | 22.78 | 37.95 | 1 |
| llama-3.2-1b-instruct-q8_0 | 18.29 | 33.96 | 20.08 | 273.88 | 3 |
| Llama-3.2-1B-Instruct.Q6_K | 15.62 | 203.16 | 15.62 | 203.16 | 1 |
| DeepSeek-R1-Distill-Qwen-1.5B-IQ4_XS | 11.13 | 23.59 | 11.13 | 23.59 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 9.08 | 57.13 | 15.29 | 140.38 | 4 |
| Gemmasutra-Mini-2B-v1-Q5_K_S | 4.99 | 9.58 | 4.99 | 9.58 | 1 |
| model-Q3_K_M | 4.05 | 40.40 | 4.05 | 40.40 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 3.63 | 9.33 | 3.63 | 9.33 | 1 |
| Llama-3.2-3B-Instruct-uncensored-Q4_0 | 0.37 | 9.26 | 0.37 | 9.26 | 1 |
Head-to-Head Record
1–50 of 115 rows
1 / 3
Performance by App Version
ImprovedRegressed