OpenCode
SST · ai coding agents · Source: data snapshot · data as of 2026-09-16
Open-source, 2nd-gen terminal agent — provider-agnostic with AST-aware context and native prompt caching.
ToolScore breakdown — why 80/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
✅ What it does well
- ▸2nd-gen token architecture: ~1–3k system prompt + AST file reading = a $20 budget lasts all day
- ▸Provider-agnostic from day one — no lock-in, swap per task
- ▸Native prompt caching slashes the cost of repeated context
- ▸Terminal + TUI that feels closer to a product than a script
⚠️ Where it costs you
- ▸Young project — pace of change is fast, occasional rough edges
- ▸Community support only; documentation still catching up with features
- ▸Fewer polished integrations than the commercial incumbents
Hidden traps — the stuff the pricing page doesn't say
Weekly-ish releases mean an upgrade can change default models or behaviors mid-project. Pin your version for long tasks.
Know the limits before you pay: read the traps above, then decide with eyes open.
Visit official site →Affiliate link where available — never affects the score.
- official OpenCode — official site (snapshot, unverified) unverified
- official GitHub repo (architecture, license) unverified
Data as of 2026-09-16. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
Google's open-source terminal agent with the most generous free tier in the category and a 1M-token
The open-source IDE assistant layer — autocomplete, chat, and agents over any model you point it at.
Open-source VS Code agent with BYOK transparency — powerful, but famous for eating tokens by design.
OpenCode review
In-depth analysis · last updated 2026-09-16
The open-source answer to the $20 burn problem
OpenCode is what happens when an open-source team takes the 2nd-generation agent architecture seriously: a small system prompt (~1–3k tokens instead of 15–25k), AST-aware file reading via tree-sitter so the agent reads only what's relevant, and native prompt caching so repeated context costs a fraction of full price.
The token math
The practical consequence: where a 1st-gen agent exhausts a $20 plan in 2–3 hours of active work, the same budget on OpenCode + your own key typically covers a full working day or more. Across a month, that's the difference between ~$60–90 (subscription top-ups) and ~$15–25 (raw tokens via a discount provider).
Provider freedom as a feature
OpenCode treats the model backend as a config choice, not an identity. OpenAI today, Anthropic tomorrow, a local model on the train — same workflow, same key management, no export ritual. That freedom is also its privacy story: nothing about your code needs to touch a vendor's training pipeline unless you point it there.
The honest downside
It's a young project moving fast. Documentation trails features, and support means Discord and GitHub issues, not a vendor SLA. Stability is good-not-great; long sessions occasionally surface an unpolished edge.
Verdict
For developers who want 2nd-gen economics without a subscription, OpenCode is the strongest new entrant in this hub — 80/100, Best Value. Use it with a pinned version and a provider-side spend cap.