family member

Qwen 3 VL-4B

1 build of Qwen 3 VL-4B measured on real machines — different packagers, precisions and engines, each a different file with its own numbers. The best each engine proved is first; every build has its own page with the machines and signatures behind it.

on ollama · 1 build

trained 256K · proved 128K context proved by Qwen F16

12.2 tok/s peak, by Qwen F16

proved by at least one build: Tool calling · Faithful to tool results · Structured output · Sustained output · Vision · Counts in images · Multiple images at once · Spatial reasoning · Writes code · Follows instructions

buildpublisherprecisionenginewhose workcontext provedbest tok/smachines
hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:F16QwenF16ollamavendor1,31,07212.21