OpenCode itself is free and open source (MIT) — you never pay for the agent. What you pay for is the model behind it, and there are four ways: free models on OpenCode Zen ($0, rotating open-weight list), OpenCode Go ($10/mo flat, 14 open-weight models with usage caps), Zen pay-as-you-go (zero markup on list prices), or an outside subscription — ChatGPT Plus via Codex OAuth, a Chinese coding plan, or a flat-rate routed API like Standard Compute from $19/mo.
| Option | Price | Models | The catch |
|---|---|---|---|
| Zen free models | $0 | ~7 rotating open-weight models | List rotates; quality varies; rate-limited |
| OpenCode Go | $10/mo flat | 14 open-weight models (Kimi, GLM, DeepSeek…) | Usage caps (5-hour/weekly/monthly metering reported); no closed frontier models |
| Zen pay-as-you-go | List price, zero markup | Frontier + open (Claude Sonnet 5 $2/$10 per MTok, GPT-5.4 $2.50/$15…) | Agent loops make per-token bills spike — users report $70 weekends |
| Flat-rate routed API | From $19/mo | Frontier + open via smart routing | Router picks the model; you do not pin versions |
The recurring community answer (including a 99-upvote thread asking exactly this) is a split setup: something flat for volume work, something frontier for hard problems. Popular combinations are Codex OAuth on a $20 ChatGPT Plus plan, GLM or Kimi coding plans in the $18–20 range, OpenCode Go at $10 for open models, and flat routed APIs for people who want frontier quality without a meter.
The recurring community complaint is plans changing underneath you — quotas rewritten, tiers killed, caps added mid-subscription. Whatever you buy, check the limit mechanics, not just the headline price.
Every cheap option meters something: Go has usage windows, free models rotate and throttle, PAYG has no cap at all (which is its own risk — the meter runs while your agent loops). The question is not which plan is cheapest, it is which limit you can live with.
Standard Compute is the flat-rate option OpenCode users pick when they want frontier models without a meter: set the base URL to api.stdcmpt.com/v1, model to standardcompute, and smart routing serves Claude, GPT, Gemini-class and efficient open models under one fixed monthly budget from $19/mo — no 5-hour windows, no weekly caps. OpenCode is our single biggest user community, which is why the setup guide is two minutes.
Yes — the agent is MIT-licensed, free forever, no account required. The model behind it is the only cost, and even that can be $0 if you use the free models on OpenCode Zen.
For open-weight models, OpenCode Go at $10/mo is the cheapest flat option. For frontier models, the community picks are ChatGPT Plus ($20, via Codex OAuth), a GLM/Kimi coding plan ($18–20), or a flat-rate routed API from $19/mo. Heavy users usually run one flat plan for volume plus PAYG for edge cases.
No plan is literally unlimited — Go has caps, coding plans have session/weekly windows, PAYG is bounded by your wallet. The closest is a fixed monthly compute budget with no per-request windows, which is the flat-rate routed model.
OpenCode's first-party model gateway: pay-as-you-go at list prices with zero markup (card fees passed through at cost), a $20 starting balance, and a rotating set of free models. It is the no-commitment way to run OpenCode on real models.