Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPhone 15 Pro

iOS

Rank

#22

Rating

1,916

±16 RD

Win Rate

90.1%

Conservative Rating

1,884

TG Rating

1,883

PP Rating

1,929

Matches

1,102

Record

993W – 109L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
EXAONE-3.5-7.8B-Instruct-Q4_K_M2442.14116609.272442.14116609.271
Qwen3-8B-Q4_K_M1336.48120381.791336.48120381.791
gemma-3-270m-it-Q2_K121.30303.37121.30303.371
google_gemma-3-270m-it-Q6_K118.31415.20118.31415.201
SmolLM2-135M-Instruct-Q2_K94.59194.8094.59194.801
smollm2-360m-instruct-q8_074.361708.3374.601709.752
Vintern-1B-v3_5-Q8_069.91169.2769.91169.271
gemma-3-270m-it-F1669.784194.5070.074199.962
Qwen3-0.6B.Q4_K_M64.04109.6464.04109.641
qwen2.5-0.5b-instruct-q6_k62.831347.4563.761397.382
qwen2.5-0.5b-instruct-q4_062.29111.7462.29111.741
Qwen3-0.6B-Q6_K52.20839.3364.021037.392
Meta-Llama-3-8B-Instruct-Q4_K_M51.85323.1151.85323.111
tinyllama-1.1b-chat-v1.0.Q4_051.22470.3460.67663.536
Meta-Llama-3-8B-Instruct.Q4_K_M45.33455.2357.25599.092
qwen2.5-1.5b-instruct-q2_k41.04452.4041.04452.401
agentica-org_DeepScaleR-1.5B-Preview-IQ4_XS39.28461.1739.28461.171
QwQ-0.5B-Distilled-SFT-f1637.691492.7837.691492.781
gemma-3-1B-it-QAT-Q4_037.55356.1340.54657.422
gemma-3-1b-it-Q8_035.51829.2835.51829.281
Qwen3-1.7B-Q4_K_M33.12401.8133.12401.811
llama-3.2-1b-instruct-q8_031.10200.9532.26647.294
DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_S29.81381.4929.81381.491
Qwen3-1.7B.Q5_K_M29.03315.1229.03315.121
DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M26.9741.8026.9741.801
DeepSeek-R1-Distill-Qwen-1.5B-Q8_026.61503.8426.61503.841
EXAONE-3.5-2.4B-Instruct-Q4_K_M25.57172.1625.57172.161
DeepSeek-R1-Distill-Qwen-1.5B-Q8_025.00488.6225.06530.302
Deepseek-Qwen2.5-1.5B-Redistil.Q8_024.64517.9224.64517.921
SmolVLM2-2.2B-Instruct-Q4_K_M23.9534.7223.9534.721
Qwen3.5-2B-UD-IQ2_XXS23.25331.3623.25331.361
gemma-3n-E2B-it-Q4_022.59237.1922.59237.191
qwen2.5-1.5b-instruct-q8_022.5647.5925.1351.174
deepseek-r1-distill-qwen-1.5b-q8_021.60306.9621.60306.961
DeepSeek-R1-Distill-Qwen-1.5B-Q6_K21.4831.6421.4831.641
gemma-3n-E2B-it-Q4_K_M21.48221.6421.62223.753
DeepSeek-R1-Distill-Qwen-1.5B-Q8_020.91135.7524.76222.272
gemma-2-2b-it.Q5_K_M18.71249.3918.71249.391
gemma-3-4b-it-q4_0_s17.51195.1617.51195.161
gemma-3n-E2B-it-Q5_K_M16.59172.2116.94181.942
Phi-3.5-mini-instruct.Q4_K_M15.68145.0016.54163.837
SmolLM2-1.7B-Instruct-Q8_015.3428.4222.22350.614
Qwen3-4B-Q4_K_M14.78141.4516.39160.052
Phi-4-mini-instruct-Q5_K_M14.59185.5414.59185.541
DeepSeek-R1-Distill-Qwen-1.5B-f1613.76491.7913.76491.791
DeepSeek-R1-Distill-Qwen-1.5B-IQ2_M13.7518.8513.7518.851
agentica-org_DeepScaleR-1.5B-Preview-f1613.75444.5813.75444.581
gemma-3-4b-it-abliterated-v2.q6_k13.64175.2113.64175.211
gemma-2-2b-it-Q6_K13.2222.5519.57306.5211
gemma-3-4b-it.Q4_K_M12.30127.6412.30127.641

150 of 92 rows

1 / 2

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With