Every open model a machine of this kind has run on the network, at its fastest build, fastest first. Speed is tokens a second, measured and signed by the machine itself.
| Model | Fastest build here | tok/s | Context | $ / M tokens |
|---|---|---|---|---|
| DeepSeek-OCR 2 Base | Q4_K_M (4-bit) by sabafallah | 412 | — | $0.63 |
| MiniCPM5 1B | Q4_K_M (4-bit) by openbmb | 381 | 4K | $0.67 |
| Gemma 3 270M | Q4_K_M (4-bit) by unsloth | 286 | — | $0.90 |
| Qwen2.5 0.5B Instruct | Q4_K_M (4-bit) by Qwen | 248 | 8K | $1.04 |
| Hunyuan MT 2 1.8B | Q4_K_M (4-bit) by tencent | 222 | — | $1.16 |
| Phi 4 Mini | Q4_K_M (4-bit) by maziyarpanahi | 205 | 4K | $1.25 |
| Gemma 4 E2B | IQ4_XS (4-bit) by unsloth | 183 | 4K | $1.41 |
| Llama 3.2 3B | Q4_K_M (4-bit) by bartowski | 99 | 64K | $2.58 |
| Sarvam-1 | Q4_K_M (4-bit) by bartowski | 20 | 4K | $13.03 |