Leaderboard
On-device LLM performance rankings powered by Glicko-2
iPhone 11
iOSRank
#100
Rating
1,677
±40 RD
Win Rate
67.3%
Conservative Rating
1,597
TG Rating
1,658
PP Rating
1,770
Matches
165
Record
111W – 54L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| SmolLM2-135M-Instruct-Q3_K_M | 99.50 | 220.39 | 99.50 | 220.39 | 1 |
| SmolLM2-135M-Instruct-Q4_K_M | 93.32 | 214.36 | 93.32 | 214.36 | 1 |
| SmolLM2-135M-Instruct-Q5_K_M | 84.77 | 195.05 | 84.77 | 195.05 | 1 |
| SmolLM2-135M-Instruct-Q2_K | 83.34 | 207.73 | 83.34 | 207.73 | 1 |
| SmolLM2-135M-Instruct-Q8_0 | 62.46 | 91.48 | 62.46 | 91.48 | 1 |
| SmolLM2-135M-Instruct-Q6_K | 56.89 | 86.57 | 56.89 | 86.57 | 1 |
| SmolLM2-135M-Instruct-F16 | 49.18 | 182.84 | 49.18 | 182.84 | 1 |
| gemma-3-270m-it-F16 | 40.33 | 164.69 | 48.30 | 173.50 | 2 |
| Qwen2-500M-Instruct-Q5_K_M | 28.57 | 291.31 | 28.57 | 291.31 | 1 |
| Qwen3-0.6B.Q4_K_M | 24.75 | 118.48 | 24.75 | 118.48 | 1 |
| tinyllama-1.1b-chat-v1.0.Q2_K | 17.07 | 14.63 | 17.07 | 14.63 | 1 |
| llama-3.2-1b-instruct-q8_0 | 10.64 | 16.42 | 13.23 | 19.85 | 2 |
| google_gemma-3-270m-it-bf16 | 9.76 | 28.63 | 9.76 | 28.63 | 1 |
| HY-MT1.5-1.8B-Q4_K_M | 9.50 | 51.73 | 9.50 | 51.73 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 8.94 | 12.20 | 8.94 | 12.20 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 8.28 | 29.31 | 17.45 | 88.75 | 5 |
| agentica-org_DeepScaleR-1.5B-Preview-Q5_K_L | 6.95 | 38.42 | 6.95 | 38.42 | 1 |
| gemmasutra-mini-2b-v1-iq4_nl-imat | 4.67 | 11.29 | 4.67 | 11.29 | 1 |
| Qwen3-4B-presinq-Q3_K_S | 1.22 | 16.49 | 1.22 | 16.49 | 1 |
Head-to-Head Record
1–50 of 110 rows
1 / 3
Performance by App Version
ImprovedRegressed