Leaderboard
On-device LLM performance rankings powered by Glicko-2
Galaxy S20 FE
AndroidRank
#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
| Model | TG Median (tok/s) | PP Median (tok/s) | TG Best | PP Best | Runs |
|---|---|---|---|---|---|
| arco-chat-merged-3.Q8_0 | 23.15 | 77.99 | 23.15 | 77.99 | 1 |
| reader-lm-0.5b-Q4_K_L | 19.42 | 63.13 | 19.42 | 63.13 | 1 |
| isk_snowflake_gguf_model | 13.05 | 33.47 | 13.05 | 33.47 | 1 |
| agentica-org_DeepScaleR-1.5B-Preview-Q4_K_L | 12.56 | 27.42 | 12.56 | 27.42 | 1 |
| qwen2.5-0.5b-instruct-q4_0 | 12.03 | 101.15 | 12.03 | 101.15 | 1 |
| ReaderLM-v2.IQ4_XS | 12.00 | 22.57 | 12.00 | 22.57 | 1 |
| llama-3.2-1b-instruct-q8_0 | 11.23 | 37.15 | 12.91 | 45.38 | 4 |
| Yi-Coder-1.5B-Chat-Q8_0 | 9.93 | 31.22 | 9.93 | 31.22 | 1 |
| qwen2.5-1.5b-instruct-q8_0 | 9.22 | 31.47 | 9.22 | 31.47 | 1 |
| qwen2.5-0.5b-instruct-fp16 | 8.36 | 33.03 | 8.36 | 33.03 | 1 |
| Qwen2.5-3B-Instruct-IQ4_XS | 7.58 | 11.44 | 7.58 | 11.44 | 1 |
| SmolLM2-1.7B-Instruct-Q8_0 | 7.40 | 21.99 | 7.92 | 21.99 | 2 |
| Qwen2.5-Coder-3B-Instruct-Q4_K_L | 6.85 | 13.95 | 6.85 | 13.95 | 1 |
| Phi-4-mini-instruct-Q4_K_M | 6.59 | 11.04 | 6.59 | 11.04 | 1 |
| Phi-3.5-mini-instruct.Q4_K_M | 5.97 | 9.45 | 5.97 | 9.45 | 1 |
| gemma-3-4b-it-Q4_K_M | 5.79 | 13.24 | 5.79 | 13.24 | 1 |
| Gemmasutra-Mini-2B-v1-Q6_K | 5.73 | 14.70 | 5.73 | 14.70 | 1 |
| gemma-2-2b-it-Q6_K | 5.39 | 11.63 | 5.53 | 11.72 | 2 |
| Qwen2.5-Coder-3B-Instruct-Q8_0 | 5.04 | 13.60 | 5.13 | 14.01 | 2 |
| Llama-3.2-3B-Instruct-Q6_K | 4.99 | 9.15 | 5.19 | 9.62 | 2 |
| Qwen2.5-Coder-7B-Instruct-Q4_K_S | 3.85 | 7.01 | 3.85 | 7.01 | 1 |
| Qwen2.5-Coder-7B-Instruct-Q3_K_L | 2.98 | 4.23 | 2.98 | 4.23 | 1 |
| Qwen2.5-Coder-7B-Instruct-Q3_K_XL | 2.48 | 4.51 | 2.48 | 4.51 | 1 |
| phi-4-mini-iq4_xs | 2.38 | 6.64 | 2.38 | 6.64 | 1 |
Head-to-Head Record
1–50 of 279 rows
1 / 6
Performance by App Version
ImprovedRegressed