Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPhone 12 Pro Max

iOS

Rank

#36

Rating

1,863

±21 RD

Win Rate

85.1%

Conservative Rating

1,821

TG Rating

1,804

PP Rating

1,894

Matches

597

Record

508W – 89L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
DeepSeek-R1-Distill-Qwen-1.5B-uncensored.f1664.871341.0064.871341.001
granite-3.1-3b-a800m-instruct-Q4_K_M27.509.8927.509.891
gemma-3-1b-it.Q4_K_S25.4038.4125.4038.411
chatgpt-5-q8_024.80480.6542.32657.802
Qwen3-1.7B-Q4_K_M22.51180.5222.51180.521
llama-3.2-1b-instruct-q8_022.2148.0822.59290.424
qwen2.5-1.5b-instruct-q8_016.92115.5217.68198.492
gemma-3-1b-it.Q2_K16.1828.5916.1828.591
tinyswallow-1.5b-instruct-q5_k_m14.8621.1614.8621.161
gemma-2-2b-it.Q4_K_M14.01126.1114.01126.111
gemma-2-2b-it-Q6_K12.81126.0012.81126.001
DeepSeek-R1-Distill-Qwen-1.5B-IQ4_NL12.7418.9612.7418.961
qwen2.5-3b-instruct-q4_k_m12.6790.1312.6790.131
DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_M12.5920.0412.5920.041
DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M12.2516.8212.2516.821
llama-3.2-3b-instruct-abliterated-q4_k_m7.0162.787.0162.781
Phi-3.5-mini-instruct.Q4_K_M4.917.844.917.841
Qwen3-1.7B-UD-Q4_K_XL4.7093.544.7093.541
Llama-3.2-3B-Instruct-Q6_K4.648.104.648.101
gemma-3-4B-it-QAT-Q4_04.256.424.256.421
qwen2.5-3b-instruct-q5_k_m3.989.523.989.521

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With