vLLM adds Gemma 4 image and audio placeholder support
October 22, 2026
The vLLM inference runtime has merged a pull request adding support for Gemma 4 image and audio placeholder strings. This update enables the runtime to handle multimodal placeholder tokens for the Gemma 4 model family.
HOW THIS AFFECTS YOU
●
builderYou can now run multimodal Gemma 4 workloads using the vLLM inference engine.