OpenCode supports zero-retention for all Go models

• OpenCode said all models in its Go subscription now run under zero-data-retention agreements, adding that user data is "not used for training" according to the announcement.
• A follow-up from OpenCode's founder said Go users are routed through "the exact same providers" they would hit directly, pushing back on claims that the $10 plan gets a degraded path or different model behavior the clarification.
• The pricing discussion matters because OpenCode says a subscription only works if it can turn a $10 plan into far more inference through provider deals; in the pricing thread, the company said a higher tier would magnify losses when those assumptions miss.
• That lands in a community context where the Hacker News launch thread praised OpenCode's multi-backend and local-model workflows but also raised concerns about telemetry and a WebUI that proxies through OpenCode infrastructure the HN thread.

The concrete product change is narrow but important: OpenCode says it has signed zero-data-retention agreements "with all providers for Go," so every model on the hosted Go plan now follows that policy announcement. That is stronger than a generic privacy claim because it describes provider-side retention terms, not just OpenCode's own handling.

The change also tightens a tension already visible in OpenCode's launch materials. Its product page describes the tool as "privacy-first" and says it stores no code or context, while the Hacker News discussion around the launch showed engineers using it across terminal, IDE, desktop, and local-model setups. Zero retention for Go makes the hosted subscription more consistent with that positioning, even though it does not by itself answer every question about the full request path.

OpenCode addressed one of the louder user suspicions directly: according to thdxr's post, the $10 Go plan uses "the exact same providers you use when you go direct." The caveat was operational rather than architectural — providers "are constantly tweaking things" and sometimes introduce small bugs — but the claim is that OpenCode is not serving a cheaper or specially degraded model variant behind the subscription.

That clarification matters because the community discussion was not only about model quality. In the Hacker News thread, users highlighted practical upside in opencode serve, multi-backend WebUI control, LSP integration, and custom IPC plugins, but others warned that telemetry still reaches OpenCode's servers and that the WebUI proxies via OpenCode infrastructure rather than running fully local. OpenCode's pricing explanation adds another piece: the company says subscriptions depend on negotiated inference economics, where a $10 plan might target something like "$40 of inference," and larger tiers are risky because any pricing error scales with them. Put together, the new message is clearer on provider retention and routing parity, while the older deployment questions around telemetry and proxying remain separate concerns.