opencode

Free (Open Source)

• MIT licensed agent, the entire thing, no feature gates
• Two-agent flow: a read-only plan agent and a separate build agent
• LSP diagnostics pulled in during editing
• Any provider: Anthropic, OpenAI, Google, GitHub Copilot, OpenRouter or a local model
• Optional client/server split so the agent runs on one box and you drive it from another
• Some Zen models are currently free while the team gathers feedback, though they are explicitly time-limited

Paid

Nothing for the agent. Anomaly sells opencode Zen, a curated model gateway:

• Pay per token, no subscription and no seat fee. Add credits and spend them
• GPT 5.6 Luna: $0.20 per 1M input, $1.20 per 1M output at 272K context or below. $0.40 and $1.80 above it. Most models are banded this way
• Claude Opus 5: $5.00 per 1M input, $25.00 per 1M output
• GPT 5.5 Pro: $30.00 per 1M input, $180.00 per 1M output
• Auto-reload adds $20 when the balance falls below $5, and both thresholds are adjustable
• Workspace-level monthly spending limits
• Card processing fees of 4.4% plus $0.30 per transaction are passed through at cost
• A separate enterprise track exists, priced on request

Self-Hosted Costs

$0 for the software. It is a CLI and TUI on your own machine, no server and no GPU required. The only bill is inference, and running a local model through Ollama or similar takes even that to zero.

When to Pay

Never for the agent. Use Zen when you want models pre-vetted for coding-agent behavior and a single balance instead of separate provider accounts, and note the 4.4% card fee makes it slightly more expensive than going direct. If you already hold an Anthropic or OpenAI key, keep using it. Compare against Cursor and GitHub Copilot at roughly $20 per seat per month: opencode plus metered tokens usually costs less for light use and more for heavy use.