I use a Claude Pro subscription for orchestration with Opus 5.5, and for everything else, I use OpenCode Go and Command Code GOAT. It’s good to have a variety of models; cognitive diversity is useful for agentic patterns involving adversarial validation. I’ve compiled the agentic patterns I use here: https://graphs.univerlab.org/patterns.

In total, I spend 40 USD. I also take advantage of all the available free tiers: OpenCode, Kiro, Codex, Antigravity, Kilo, Vibe, Cursor, CodeBuddy, Devin, and FreeBuff.

That’s for my personal use; for work, I’m provided with 10K AI credits for GitHub Copilot and Codex Pro.