Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPhone 17 Air

iOS

Rank

#11

Rating

1,972

±21 RD

Win Rate

95.6%

Conservative Rating

1,929

TG Rating

1,972

PP Rating

1,961

Matches

590

Record

564W – 26L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
Qwen3-0.6B-Q4_K_M115.261541.37115.261541.371
Qwen3-0.6B-Q8_082.821551.9982.821551.991
qwen2.5-1.5b-instruct-q8_036.36598.4836.40599.573
gemma-3n-E2B-it-Q4_K_M31.29234.6031.29234.601
gemma-2-2b-it-Q6_K26.26349.8726.26349.871
qwen2.5-3b-instruct-q5_k_m24.77255.1225.05258.204
moondream2-text-model-f16_ct-vicuna22.28645.3222.28645.321
LFM2.5-VL-1.6B-BF1622.19732.8822.19732.881
Qwen3-4B-Instruct-2507-Q4_K_M21.69202.4221.69202.421
Qwen3VL-4B-Instruct-Q4_K_M21.51201.5621.51201.561
Qwen3.5-4B-Q4_K_M12.07159.9812.07159.981
MechaEpstein-8000.Q4_K_M10.34102.5010.34102.501
Qwen3.5-4B-Q6_K9.23146.239.23146.231
Qwen2.5-7B-Instruct-Q5_K_M8.9093.188.9093.181
Qwen3-4B-Instruct-2507-IQ4_XS8.7312.948.7312.941
Qwen3.5-9B-UD-IQ2_M8.6793.888.6793.881
Qwen3.5-4B-Q6_K8.35130.938.35130.931
Meta-Llama-3.1-8B-Instruct-Q5_K_M7.6380.357.6380.351

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With