Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPad Air 11 inch 6th Gen

iOS

Rank

#7

Rating

1,992

±20 RD

Win Rate

97.6%

Conservative Rating

1,952

TG Rating

1,994

PP Rating

1,985

Matches

658

Record

642W – 16L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
DeepSeek-R1-Distill-Llama-8B-Q4_K_M116.761050.32116.761050.321
DeepSeek-R1-0528-Qwen3-8B-UD-Q4_K_XL108.371328.93108.371328.931
gemma-3-1B-it-QAT-Q4_075.37965.7286.051441.033
Qwen3-0.6B.Q6_K63.86669.7363.86669.731
Qwen3-1.7B.Q4_K_M52.95207.7352.95207.731
tinyswallow-1.5b-instruct-q8_047.42813.8347.42813.831
DeepSeek-R1-Distill-Qwen-1.5B-Q8_034.1383.2634.1383.261
Qwen2.5-1.5B-Instruct.Q8_034.0489.7134.0489.711
Llama-3.2-1B-Instruct.IQ1_M27.8140.1027.8140.101
qwen2.5-3b-instruct-q5_k_m27.31330.6830.76344.604
gemma-3-4B-it-QAT-Q4_022.8152.9522.8152.951
Phi-3.5-mini-instruct.Q4_K_M21.11192.8230.79254.452
Gemmasutra-Mini-2B-v1-Q6_K18.9633.3318.9633.331
gemma-2-2b-it-Q6_K18.7636.6034.14509.353
DeepSeek-R1-Distill-Llama-8B-Q2_K18.62151.3418.62151.341
Hermes-3-Llama-3.2-3B.Q6_K13.5822.5913.5822.591
DeepSeek-R1-Distill-Qwen-7B-IQ3_M13.18132.2013.18132.201
oh-dcft-v3.1-claude-3-5-haiku-20241022.Q3_K_S12.70138.5012.70138.501
DeepSeek-R1-Distill-Qwen-7B-Q4_K_M12.40157.9612.40157.961

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With