Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPad Pro 12.9 inch 5th Gen
iOSRank
#13
Rating
1,960
±21 RD
Win Rate
94.4%
Conservative Rating
1,918
TG Rating
1,953
PP Rating
1,975
Matches
628
Record
593W – 35L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| llama-3.2-1b-instruct-q8_0 | 37.91 | 106.70 | 37.91 | 106.70 | 1 |
| HY-MT1.5-1.8B-Q8_0 | 29.89 | 567.06 | 29.89 | 567.06 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 25.46 | 72.57 | 25.46 | 72.57 | 1 |
| gemma-2-2b-it-Q6_K | 20.86 | 193.98 | 23.75 | 355.97 | 2 |
| Meta-Llama-3.1-8B-Instruct-Q2_K | 15.77 | 126.25 | 15.77 | 126.25 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 15.33 | 22.65 | 21.35 | 258.41 | 3 |
| Llama-3.2-3B-Instruct-Q6_K | 14.12 | 132.99 | 14.29 | 244.33 | 2 |
| Phi-3.5-mini-instruct.Q4_K_M | 14.11 | 19.44 | 14.11 | 19.44 | 1 |
| DeepSeek-R1-Distill-Qwen-7B-Q4_K_S | 12.64 | 116.37 | 12.64 | 116.37 | 1 |
| Qwen3.5-4B-IQ4_NL | 10.21 | 152.30 | 10.21 | 152.30 | 1 |
| deepseek-r1-distill-qwen-7b-abliterated-v2-Q4_K_M | 8.13 | 12.09 | 8.13 | 12.09 | 1 |
| Llama3-8B-1.58-100B-tokens-TQ1_0-F16 | 7.32 | 10.90 | 7.32 | 10.90 | 1 |
| DeepSeek-R1-Distill-Llama-8B-Abliterated.Q4_K_M | 4.48 | 10.11 | 5.18 | 10.23 | 2 |
Head-to-Head Record
1–50 of 287 rows
1 / 6
Performance by App Version
ImprovedRegressed