Preserve thinking
Haiku requests keep manual thinking enabled, with a bounded budget that respects `max_tokens`.
A focused plugin that fixes Haiku's "You're out of extra usage" failure in OpenCode through Sub2API, while preserving thinking and leaving Sub2API, OpenCode, and OMO code untouched.
Narrow fix
Opus and Sonnet already passed through the route. Haiku failed because its request fingerprint diverged: incomplete system body, missing Claude Code metadata, and thinking shape drift.
Haiku requests keep manual thinking enabled, with a bounded budget that respects `max_tokens`.
Provider and body checks exit early for non-Haiku models, leaving Sonnet and Opus untouched.
The fix lives in an OpenCode plugin, so Sub2API, OpenCode core, and OMO can keep updating normally.
Request surgery
The plugin does not impersonate an account or carry credentials. It normalizes the request OpenCode is already sending through your configured provider.
Install
Add it to OpenCode's plugin list. Existing Sub2API auth and provider configuration stay exactly where they are.
{
"plugin": [
"opencode-sub2api-haiku-fix@latest"
]
}