Gemma Playground: Parallel Agents in Action
Gemma 4 runs 10 parallel local sub-agents at 170+ tokens/sec on one machine.
“This is the new standard for local parallel AI.”
A Google I/O demo showcased Gemma 4's 26B model orchestrating 10 parallel sub-agents locally to generate SVG art, claiming over 170 tokens/sec across concurrent tasks on a single machine. It positions on-device multi-agent batching as viable for enterprise task decomposition and private office-wide chatbots on one or two GPUs. The signal matters as a marketing push for local, private parallel AI, though it reads as a promotional demo rather than independently verified capability.