[Bug] llm.discoverModels returns build-time bundled catalog for known providers — newly released models (e.g. x-preview-f-free) never appear via "Fetch Models" #4564
Replies: 1 comment
|
根因定位准确( 1. 你自己找到的那条就是当前最好的解法。 手工写显式 2. 这个短路是「已知 provider 用打包目录」的设计选择,不是漏判。 所以它的后果是系统性的:任何内置 provider 的新模型都要等 DSH 发版带新目录。相关的还有 #3672(rc.8 把 3. 如果你想让 opencode 的目录自动跟随,社区有个专门做这条的 provider 包: dsh plugin --profile web add pi2dsh
dsh plugin --profile web add pi-opencode-go-provider披露清楚: 边界:这条路是绕开 |
Uh oh!
There was an error while loading. Please reload this page.
Environment
@deepseek-ai/dsh)@earendil-works/pi-ai0.82.1npm i -g @deepseek-ai/dshSteps to reproduce
opencode-go/opencodewith onlyapiKeyEnvset (no explicit models).x-preview-f-free("Ox Alpha Free") — does not appear.
models:entry for the same provider worksimmediately, proving the endpoint itself serves the new model.
Actual behavior
Discovery returns only a static, build-time bundled catalog and never
interrogates the live endpoint.
Root cause (traced through the installed bundle):
@deepseek-ai/dsh-llm-pi-ai/lib/index.js—discoverModels():All reactions