Leaderboard
On-device LLM performance rankings powered by Glicko-2
Galaxy S22 Ultra
AndroidRank
#123
Rating
1,535
±14 RD
Win Rate
53.4%
Conservative Rating
1,507
TG Rating
1,550
PP Rating
1,402
Matches
1,316
Record
703W – 613L
Models Tested
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| llama-3.2-1b-instruct-q8_0 | 11.02 | 52.56 | 15.82 | 67.84 | 4 |
| qwen2.5-1.5b-instruct-q8_0 | 8.20 | 33.94 | 12.11 | 35.48 | 2 |
| SmolLM2-1.7B-Instruct-Q8_0 | 7.88 | 29.80 | 11.74 | 32.79 | 3 |
| Phi-3.5-mini-instruct.Q4_K_M | 5.91 | 10.41 | 8.44 | 13.62 | 6 |
| DeepSeek-R1-Distill-Qwen-1.5B-IQ4_XS | 5.88 | 12.30 | 5.88 | 12.30 | 1 |
| gemma-3n-E2B-it-Q4_K_M | 5.58 | 16.20 | 5.58 | 16.20 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 5.58 | 13.52 | 5.58 | 13.52 | 1 |
| gemma-2-2b-it-Q6_K | 5.36 | 12.82 | 6.79 | 19.36 | 6 |
| Llama-3.2-3B-Instruct-Q6_K | 4.98 | 9.68 | 5.55 | 18.18 | 6 |
| Dark-Hermes3-Llama3.2-3B.Q2_K | 4.56 | 9.38 | 4.56 | 9.38 | 1 |
| qwen2.5-3b-instruct-q5_k_m | 2.79 | 7.52 | 3.81 | 9.37 | 2 |
| DeepSeek-R1-Distill-Llama-8B-Q3_K_L | 2.64 | 3.93 | 2.64 | 3.93 | 1 |
| DeepSeek-R1-Distill-Llama-8B-Q2_K | 2.63 | 3.83 | 2.63 | 3.83 | 1 |
| gemma-3-4b-it.Q8_0 | 2.51 | 13.40 | 2.51 | 13.40 | 1 |
Head-to-Head Record
1–50 of 316 rows
1 / 7
Performance by App Version
ImprovedRegressed