Inference · compare

DeepSeek-OCR 2 Base vs Qwen2.5 0.5B Instruct.

Side by side only on machines of the same kind, so the difference is the model, not the computer. Each figure is the fastest build of that model on that machine.

2machine kinds both ran on
MachineDeepSeek-OCR 2 Base tok/sQwen2.5 0.5B Instruct tok/sContext A / B
Intel(R) Iris(R) Xe Graphics · 1 GB15Q4_K_M (4-bit) by sabafallah23Q4_K_M (4-bit) by Qwen— / 8K
GeForce RTX 3080 Ti · 12 GB412Q4_K_M (4-bit) by sabafallah248Q4_K_M (4-bit) by Qwen— / 8K