Leaderboard

On-device LLM performance rankings powered by Glicko-2

REDMAGIC 10 Pro

Android

Rank

#35

Rating

1,860

±16 RD

Win Rate

84.7%

Conservative Rating

1,829

TG Rating

1,887

PP Rating

1,825

Matches

1,099

Record

931W – 168L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
ERNIE-4.5-0.3B-PT-Q8_0119.72471.86119.72471.861
Qwen2-0.5B-Instruct-Q8_069.93489.2570.56493.562
rwkv7-goose-world2.8-0.1b-F1651.46524.3551.46524.351
olmoe-1b-7b-0924-instruct-q4_k_m45.8598.6646.1599.362
granite-3.1-3b-a800m-instruct-Q8_038.91131.2039.00131.322
llama-3.2-1b-instruct-q8_029.17192.7234.51200.024
gemma-3-1b-it.Q4_K_M28.13154.1028.13154.101
SmolLM2-1.7B-Instruct-Q8_026.05111.9126.07112.712
qwen2.5-3b-instruct-q5_k_m19.6934.2920.8634.985
gpt-oss-20b-MXFP415.6656.5615.6656.561
SmolLM2-1.7B-Instruct-Q8_014.2587.6514.2587.651
Qwen2.5-1.5B-Instruct.fp1614.2442.1014.6842.913
GLM-4.7-Flash-UD-Q4_K_XL14.1432.6014.1432.601
Phi-3.5-mini-instruct.Q4_K_M12.6117.1913.5932.505
Llama-3.2-3B-Instruct-Q6_K11.8924.9414.0333.104
gemma-2-2b-it-Q6_K11.6023.1612.4032.653
Llama3-8B-1.58-100B-tokens-TQ1_0-F1611.2018.3911.2018.391
Llama-3SOME-8B-v2-Q4_K_M9.7823.389.7823.381
DeepSeek-R1-Distill-Qwen-1.5B-f329.1634.269.3336.803
gemma-3-4b-it.Q8_08.9549.848.9549.841
GLM-4.7-Flash.q3_k_l7.6611.147.6611.141
DeepSeek-R1-Distill-Qwen-7B-Q8_07.5227.647.5233.092
gemma-3-4b-it-IQ4_NL6.6270.266.6270.261
gemma-3n-E4B-it-UD-Q8_K_XL6.4114.876.4114.871
gemma-2-9b-it.Q2_K5.447.375.447.371
gemma-3-12b-it.Q4_K_M4.678.014.678.011
Qwen2.5-7B-Instruct.Q4_K_M4.0212.714.0212.711
Qwen3-8B-f163.536.623.536.621
DeepSeek-R1-Distill-Qwen-14B-Q8_03.328.613.328.611
gpt-oss-20b-F163.329.143.329.141
Qwen2.5-7B-Instruct.fp163.126.163.126.161
Qwen2.5-14B-Instruct-Q8_02.9211.792.9211.791
gemma-3-4b-it-qat-UD-Q8_K_XL2.899.482.899.481
cognitivecomputations_Dolphin3.0-R1-Mistral-24B-Q6_K_L1.612.931.612.931
gemma-3-27b-it.Q3_K_S1.321.611.321.611

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With