With what Gemma 4 can achieve, I am seriously considering hosting the best open model (for a given time frame) on a performant VM and providing "almost" free AI access for agentic use in OpenClaw.
AI API costs are one of the biggest constraints for people who want to use personal AI agents and keep costs under control. But for this to happen, the mode needs to be very good for agentic use cases. And from how it looks, Gemma 4 promises exactly this.