Leaderboard

On-device LLM performance rankings powered by Glicko-2

Pixel 8 Pro

Android

Rank

#135

Rating

1,507

±14 RD

Win Rate

50.7%

Conservative Rating

1,479

TG Rating

1,495

PP Rating

1,630

Matches

1,401

Record

710W – 691L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
Llama-3.2-1B-Instruct.IQ4_XS31.8248.6147.9752.552
Thinker-SmolLM2-135M-Instruct-Reasoning.i1-Q4_K_M25.40218.1025.40218.101
gemma-3-270m-it-IQ4_XS20.08291.8020.08291.801
SmolLM2-360M-Instruct-Q8_014.79113.4914.79113.491
gemma-3-1B-it-QAT-Q4_012.4890.5312.4890.531
Qwen3-0.6B-IQ4_XS12.1555.3912.1555.391
Llama-3.2-1B-Instruct.Q6_K11.9340.5513.1449.583
Llama-3.2-1B-Instruct-Uncensored.Q4_K_M10.6751.0710.6751.071
Qwen2-VL-2B-Instruct-Q4_08.7456.758.7456.751
gemma-3-1b-it-Q2_K_L8.4659.118.4659.111
SmolLM2-1.7B-Instruct-Q8_08.1348.829.6664.322
DeepSeek-R1-Distill-Qwen-1.5B-Q8_08.0647.418.0647.411
llama-3.2-1b-instruct-q8_07.7152.738.7856.012
qwen2.5-1.5b-instruct-q8_07.3853.828.0257.273
Ministral-3-3B-Instruct-2512-Q4_K_M6.6026.457.5629.932
DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_M6.5922.426.5922.421
DeepSeek-R1-Distill-Qwen-1.5B-Q2_K_L6.5623.326.5623.321
DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M6.3528.716.4130.942
DeepHermes-3-Llama-3-3B-Preview-q45.6825.995.6825.991
HuggingFaceTB_SmolLM3-3B-Q4_K_M5.6716.425.6716.421
Qwen3.5-2B-Q4_K_M5.6145.515.6145.511
DeepSeek-R1-Distill-Qwen-1.5B-f165.3113.105.3113.101
DeepSeek-R1-Distill-Qwen-1.5B-Q8_05.2336.565.2336.561
qwen2.5-3b-instruct-q5_k_m5.1724.946.0726.523
Qwen3.5-2B.Q6_K5.0443.855.0443.851
huihui-ai_gemma-3-1b-it-abliterated-Q8_05.0274.195.0274.191
Llama-3.2-3B-Instruct-uncensored-Q4_K_M4.9018.764.9018.761
Gemmasutra-Mini-2B-v1-Q6_K4.8816.846.5327.903
gemma-2-2b-it-Q6_K4.7417.856.7335.679
nsfw-flash-q4_k_m4.5726.424.5726.421
gemma-3-4b-it-Q2_K4.5311.554.5311.551
gemma-2-2b-it-Q4_K_M4.4921.184.4921.181
DeepHermes-3-Llama-3-3B-Preview-q64.4210.834.4210.831
gemma-3-4b-it-Q2_K_L4.3911.154.3911.151
gemma-3-4b-it-Q4_K_M4.3514.564.3514.561
Qwen3-1.7B.Q3_K_M4.2420.694.2420.691
Qwen3.5-0.8B-BF164.2118.474.2118.471
Phi-3.5-mini-instruct.Q4_K_M4.0014.025.1816.922
gemma-3-4b-it.Q4_K_M3.5516.083.5516.081
DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored-Q4_K_M-imat3.4610.823.4610.821
gemma-3-4b-it-qat-q4_0-unquantized-q4_k_m3.4511.423.4511.421
gemma-3-4b-it-qat-Q8_03.3315.653.3315.651
Llama-3.2-3B-Instruct-uncensored-Q6_K3.1911.743.1911.741
Llama-3.2-3B-Instruct-Q6_K3.178.954.9012.893
Phi-4-mini-instruct.Q5_K_M2.918.542.918.541
gemma-3-4b-it.Q6_K2.7910.122.7910.121
DeepHermes-3-Llama-3-8B-q42.7311.242.7311.241
DeepSeek-R1-Distill-Qwen-7B-IQ2_M2.333.902.333.901
DeepSeek-R1-Distill-Qwen-7B-Q8_01.539.971.539.971
Llama-3.2-4X3B-MOE-Hell-California-10B-D_AU-Q4_k_m1.405.871.405.871

150 of 53 rows

1 / 2

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With