I care a lot about open source, picking my own model, and not being locked into one vendor’s stack. That bias shapes how I compare these two ways of running Claude Opus (and other models) against your repo and shell.
Both let you chat with the codebase and run terminal work; the difference is mostly who owns the knobs — you, or a tightly integrated product.
How I slice it: OpenCode vs. Claude Code (with Opus when applicable)
What I actually prefer
Open source — I like tools I can read, fork, and reason about.
Git-native flow: versions live in normal Git, not only in an app-specific checkpoint format.
I’ll be honest: my time in Claude Code has been great — it handles repos and multi-file edits in a way that feels effortless. I’m not trying to bring it down. So far it’s the best; I’m explaining why I still lean OpenCode for day-to-day philosophy.
When OpenCode wins for me
Hybrid cost: Opus (or similar) for hard bits, then switch down for repetitive churn.
Agents & skills as markdown/ YAML in my repo — matches how I already document habits.
Remote / server: client–server layout is handy for Docker or driving a session from another box via HTTP.
When I’d still grab Claude Code
Tight Claude integration: tool use and context often feel more predictable without me tuning YAML.
Subscription math: if you’re already on Pro/ Max, the CLI can beat pay-per-token through a generic client.
Low-friction experiments: snapshots beat “commit everything” when I’m spiking and don’t want Git noise yet.