Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPad Pro 11 inch 4th Gen

iOS

Rank

#4

Rating

2,003

±18 RD

Win Rate

98.6%

Conservative Rating

1,967

TG Rating

2,003

PP Rating

2,002

Matches

792

Record

781W – 11L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
Meta-Llama-3-8B-Instruct.Q4_K_M122.531162.42122.531162.421
Qwen1.5-0.5B-Chat-Q5_K_M81.39214.6881.39214.681
granite-3.1-3b-a800m-instruct-Q4_K_M67.7027.2667.7027.261
gemma-3-1b-it-Q8_066.341593.0766.341593.071
llama-3.2-1b-instruct-q8_053.40679.5560.731255.982
qwen2.5-1.5b-instruct-q8_041.14482.7948.48880.522
tinyswallow-1.5b-instruct-q5_k_m39.0960.2939.0960.291
gemma-2-2b-it-Q6_K35.81509.7936.04521.912
tinyswallow-1.5b-instruct-q8_034.0890.8234.0890.821
gemma-3-4b-it-IQ4_NL34.05401.9434.05401.941
gemma-3-4b-it-Q4_K_M30.33343.4330.33343.431
translategemma-4b-it.i1-Q6_K24.96303.1224.96303.121
Phi-3.5-mini-instruct.Q4_K_M21.87163.0731.75308.702
DeepSeek-R1-Distill-Qwen-1.5B-IQ2_M20.2529.9020.2529.901
DeepSeek-R1-Distill-Qwen-7B-abliterated.Q4_K_S19.91176.6019.91176.601
sarashina2.2-3b-instruct-v0.1-Q8_017.0341.2417.0341.241
gemma-3-4b-it.Q8_013.3733.2213.4833.292
Qwen3-4B-Q8_011.2525.7411.2525.741
DeepSeek-R1-Distill-Qwen-7B-IQ2_M9.8689.6816.04175.634
gemma-2-2b-it.IQ1_M9.8015.039.8015.031
Llama-3-ELYZA-JP-8B-IQ1_S4.706.474.706.471
DeepSeek-R1-Distill-Llama-8B-Q2_K_L1.385.161.385.161

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With