Leaderboard

On-device LLM performance rankings powered by Glicko-2

iPhone 15 Pro Max

iOS

Rank

#18

Rating

1,931

±15 RD

Win Rate

91.5%

Conservative Rating

1,901

TG Rating

1,887

PP Rating

1,945

Matches

1,241

Record

1136W – 105L

Models Tested

ModelTG Median (tok/s)PP Median (tok/s)TG BestPP BestRuns
llama-3.1-8b-instruct-q4_02113.5694180.112113.5694180.111
DeepSeek-R1-Distill-Qwen-14B-IQ2_XS1179.4537109.391179.4537109.391
DeepSeek-R1-Distill-Qwen-14B-IQ2_S284.2220905.89284.2220905.891
DeepSeek-R1-Distill-Qwen-1.5B-uncensored.f16120.0911392.49226.4422307.362
ruvltra-claude-code-0.5b-q4_k_m95.931298.8795.931298.871
gemma-3-270m-it-F1676.14306.0276.14306.021
Mistral-Nemo-Instruct-2407-IQ2_M61.59523.0761.59523.071
Nano-Vlm-Processor-494M-F1643.801690.5343.801690.531
DeepSeek-R1-Distill-Qwen-1.5B-Q2_K38.75425.0038.75425.001
DeepSeek-R1-Distill-Qwen-1.5B-uncensored.Q4_K_M35.56430.5935.56430.591
gemma-3-1b-it.Q8_034.73807.1334.73807.131
DeepSeek-R1-Distill-Qwen-1.5B-IQ2_M32.54338.0241.01447.092
Qwen3-VL-2B-Instruct-Q4_K_M31.64461.3231.64461.321
Llama-3.2-1B-Instruct.Q8_029.94650.5129.94650.511
llama-3.2-1b-instruct-q8_029.2759.8829.2759.881
Qwen2.5-0.5B-Instruct-Q4_K_S29.0679.5129.0679.511
qwen2.5-1.5b-instruct-q8_025.80490.5526.38550.994
Qwen3-1.7B.Q4_K_M25.54219.4532.77404.572
DeepSeek-R1-Distill-Qwen-1.5B-Q8_024.54473.8524.54473.851
tinyswallow-1.5b-instruct-q5_k_m24.5230.5424.5230.541
Dolphin3.0-Qwen2.5-3b-IQ2_M23.98202.6323.98202.631
tinyswallow-1.5b-instruct-q8_023.71326.3824.58456.846
DeepSeek-R1-Distill-Qwen-1.5B-uncensored.Q2_K22.2731.9622.2731.961
qwen2.5-3b-instruct-q4_k_m20.75216.2420.75216.241
lucy_128k-Q8_020.5241.6920.5241.691
gemma-3-1b-it.fp1620.24865.7320.24865.731
google_gemma-3-1b-it-bf1620.20620.8820.20620.881
Qwen3.5-2B-Q4_K_M19.28314.7519.28314.751
DeepSeek-R1-Distill-Qwen-1.5B-Q8_019.00244.9924.63459.784
Gemmasutra-Mini-2B-v1-Q6_K18.64254.8719.28310.214
Qwen3.5-2B-UD-Q4_K_XL18.59297.2618.59297.261
SmolLM2-1.7B-Instruct-Q8_017.4131.0717.4131.071
sarashina2.2-3b-instruct-v0.1-q4_k_m-imat16.40150.4616.40150.461
gemma-3-4b-it-Q4_K_M16.10188.8716.10188.871
qwen2.5-3b-instruct-q5_k_m15.02174.74127.9622651.1116
google.gemma-3-4b-it.Q6_K13.14173.0913.14173.091
mistral-7b-uncensored-Q2_K13.1083.4913.2983.632
gemma-3n-E2B-it-IQ4_NL12.7521.4012.7521.401
gemma-3n-E2B-it-Q5_K_M12.6121.4312.6121.431
nsfw-3b-q4_k_m12.2919.0212.2919.021
Llama-3.2-3B-Instruct-Q6_K12.25160.5916.08218.4612
Llama-3.2-3B-Instruct.Q8_012.24204.4712.24204.471
gemma-2-2b-it-Q6_K11.9561.1519.71307.7616
Qwen_Qwen3-4B-Thinking-2507-IQ4_NL11.56116.2211.56116.221
translategemma-4b-it-Q6_K11.25165.0111.25165.011
Phi-4-mini-instruct.Q6_K10.36132.8012.82158.622
gemma-3-4b-it.Q4_K_M9.8216.499.9016.592
sarashina2.2-3b-instruct-v0.1-Q8_09.5577.339.83133.942
gemma-3-4B-it-QAT-Q4_09.4816.169.4816.161
Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distill-heretic-v3-IQ4_NL9.21124.109.21124.101

150 of 75 rows

1 / 2

Head-to-Head Record

Performance by App Version

ImprovedRegressed

Compare With