Leaderboard

On-device LLM performance rankings powered by Glicko-2

Galaxy S20 FE

Android

Rank

#129

Rating

1,528

±15 RD

Win Rate

52.7%

Conservative Rating

1,497

TG Rating

1,548

PP Rating

1,376

Matches

1,121

Record

591W – 530L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
arco-chat-merged-3.Q8_023.1577.9923.1577.991
reader-lm-0.5b-Q4_K_L19.4263.1319.4263.131
isk_snowflake_gguf_model13.0533.4713.0533.471
agentica-org_DeepScaleR-1.5B-Preview-Q4_K_L12.5627.4212.5627.421
qwen2.5-0.5b-instruct-q4_012.03101.1512.03101.151
ReaderLM-v2.IQ4_XS12.0022.5712.0022.571
llama-3.2-1b-instruct-q8_011.2337.1512.9145.384
Yi-Coder-1.5B-Chat-Q8_09.9331.229.9331.221
qwen2.5-1.5b-instruct-q8_09.2231.479.2231.471
qwen2.5-0.5b-instruct-fp168.3633.038.3633.031
Qwen2.5-3B-Instruct-IQ4_XS7.5811.447.5811.441
SmolLM2-1.7B-Instruct-Q8_07.4021.997.9221.992
Qwen2.5-Coder-3B-Instruct-Q4_K_L6.8513.956.8513.951
Phi-4-mini-instruct-Q4_K_M6.5911.046.5911.041
Phi-3.5-mini-instruct.Q4_K_M5.979.455.979.451
gemma-3-4b-it-Q4_K_M5.7913.245.7913.241
Gemmasutra-Mini-2B-v1-Q6_K5.7314.705.7314.701
gemma-2-2b-it-Q6_K5.3911.635.5311.722
Qwen2.5-Coder-3B-Instruct-Q8_05.0413.605.1314.012
Llama-3.2-3B-Instruct-Q6_K4.999.155.199.622
Qwen2.5-Coder-7B-Instruct-Q4_K_S3.857.013.857.011
Qwen2.5-Coder-7B-Instruct-Q3_K_L2.984.232.984.231
Qwen2.5-Coder-7B-Instruct-Q3_K_XL2.484.512.484.511
phi-4-mini-iq4_xs2.386.642.386.641

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With