OpenCode free usage limit reached: why, and what ends it
OpenCode's "You've used today's free messages" and "Free usage limit reached": the allowance is counted per network, so a server or VPN can start spent.
Both messages mean the day’s allowance on OpenCode’s free models is used up. OpenCode’s terminal shows “You’ve used today’s free messages. They reset at …”, and opencode run stops with “Free usage limit reached. Connect a provider or subscribe to OpenCode Go to keep going, or switch models.” The allowance is counted per network, not per person or session, so a cloud server, a VPN or a shared connection can arrive with it already spent by someone else. Wait for the reset time shown, or end it now by connecting a provider with /connect (OpenCode Go, $10 a month in November 2026, or your own key) and choosing one of its models. OpenCode publishes no number for the free cap.
The wordings, and which OpenCode prints which
In OpenCode 2.3.0, the release of November 13, 2026, the terminal interface shows:
You've used today's free messages. They reset at <time>.
Upgrade to OpenCode Go → /connect · Switch model → /modelsCopiedThe reset time comes from the free service’s own answer; when it sends none, the line stops after “free messages”. A headless opencode run, and anything else that reads the session’s error, gets the second form:
Free usage limit reached. Connect a provider or subscribe to OpenCode Go to keep going, or switch models.CopiedA third, shorter wording fills the issue tracker: “Free usage exceeded, subscribe to Go”, sometimes with a retry countdown many hours long. Closing such reports, OpenCode’s maintainers say most of them come from version 1, “which no longer supports free models.” On version 1, then, waiting for the reset may not help, and reports of an allowance that never comes back are among the problems the pinned free-tier thread collects. Its first step is to get on the latest version 2. opencode --version prints 1.… or opencode v2.…, and installing OpenCode on Linux covers the move, including why the short install line still gives version 1.
Why a server can be out before its first prompt
No page of OpenCode’s says how the allowance is counted, but its source does. A comment in the 2.3.0 app reads: “Free usage is counted per network, not per session, so one refusal anywhere holds until it resets. Connecting a paid provider ends it early.” Reports describe the network as the public address OpenCode’s service sees, so everyone who reaches it from one address shares one allowance:
- A cloud server, whose outgoing address other machines may share or may have used before you.
- A VPN or proxy, whose exit address carries everyone else using it.
- A mobile carrier, an office, a school or shared housing, where many devices go out through one address.
- Your own CI or several agents on one machine, which draw on the same allowance as the person at the keyboard.
That is how a first prompt can meet the limit. The per-network count is the source’s; exactly how OpenCode groups addresses is not published. A different message means a different problem. A network OpenCode has flagged gets “OpenCode’s free models aren’t available from your current network because we’ve detected abusive traffic from this IP address. This isn’t related to your account.”, with the advice to turn off a VPN or proxy or subscribe to Go. That is a block, not the daily count, and a report of it was pointed to OpenCode’s support address, help@anoma.ly, as the only way to have it reviewed.
What ends it
The pinned thread gives the same three ways out the message does: “Wait for the reset time shown, subscribe to Go, or connect another provider with /connect.” In practice:
- Wait. The terminal shows when the allowance resets, when the service says. OpenCode’s Go page says only “Free models have a daily request limit”, and on November 14, 2026 the one number on OpenCode’s pages was LongCat 2.5 Free’s own 100 requests a day.
- Subscribe to OpenCode Go ($10 a month on November 14, 2026). Run
/connect, sign in, and pick a model whose name startsopencode-go/in/models. A subscriber still on anopencode/…-freemodel is still drawing on the free allowance and still meets the limit; the thread’s advice is to pick a Go model “rather than a free model.” - Connect another provider. An API key or a supported subscription, through
/connectoropencode auth loginin an SSH session. For the background service that unattended runs use, a provider’s key can go in its environment instead, as inopencode service set env ANTHROPIC_API_KEY …. - Switch models, to one from a provider you already have, with
/modelsoropencode run -m provider/model. Another free model is no way round it: in the source’s words, one refusal anywhere holds until it resets.
An agent that runs unattended on a server is the case the free allowance serves worst: its network may share the count, and a run that meets the limit simply fails. Giving it a model from a provider of its own, rather than a free one, is what keeps it running. What Go and the other options cost is in OpenCode’s pricing, and which free models there are, with the catches on each, in OpenCode’s models. If the message is instead “OpenCode’s free tier can only be used from within OpenCode”, the allowance is not the problem: that refusal has causes of its own.