I know I keep changing this, but I’m starting to lean towards Gemini 3 Flash (or 2.5) is the best default model for @OpenClaw. I’ve had mine running since late January, and at this point context window is starting to become an issue, even with a vector database. You absolutely could run Sonnet/Opus as a default, but that would be very expensive. You should still add every LLM API you like or have access to. Set Gemini as your default, and then whenever you add a skill or cron job, ask FIRST what would be the best LLM it has access to for the task, and it’s normally pretty good about adding the most cost efficient one.