Less than half a month, @openclaw burned $2,000. Sharing my current model setup and money-saving tips: Main Agent + three sub-Agents: claude sonnet-4.6, balanced and stable, rarely any bugs in daily use. Dedicated for strategy/analysis Agent: claude opus-4.6, expensive but worth it. 4 money-saving hacks to make Agents rely more on calls, using subscription quotas instead of their own API Tokens: 1. Code > 50 lines → Let the Agent call Codex (GPT PRO /month, barely runs out). 2. Deep thinking → GPT 5.4 PRO (uses subscription quota). 3. Tweet search/reporting → Grok (uses subscription quota). 4. Poster generation → Gemini API (Google gave $300 in credits). By the way, here’s why Gemini runs on subscription without getting banned: Agents are all using APIs, and only the above operations let Agents call the official CLI/browser interaction mode. Most people who get banned are running OAuth as an API, which Anthropic/Google explicitly prohibits. Anyone else got money-saving tips?