Repository navigation
[FEATURE]: Refresh models manually #4734
Description
Feature hasn't been suggested before.
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
the problem
afaik there is no way to manually refresh the models list. there are several issues reported that are basically saying "i can't see the latest model just released" and the devs are pointing to manually remove the cache file (~/.cache/opencode/models.json). e.g. look at #4493, #4463, #2961. this is not a good ux and wastes dev time.
the solution
i think we could just add a keyboard shortcut in the "select model" modal, named "refresh models list" with the possible default shortcut of ctrl+r which would basically just call the refresh method at ./packages/opencode/src/provider/models.ts
if this makes sense i could work on a pr
Activity
github-actions commentedon Nov 25, 2025
This issue might be a duplicate of or closely related to existing issues. Please check:
These issues all describe the same root problem: users cannot refresh the models list when it becomes stale, requiring manual cache file deletion as a workaround. Implementing the refresh shortcut proposed in this issue would resolve all of them.
rekram1-node commentedon Nov 25, 2025
Yup definitely wanna do that. Also note that you can do
opencode models --refreshnow but still yeah needs to be in the tuikristopolous commentedon Feb 6, 2026
Also it doesn't add all the models. Essentially this should be blank
but instead it's very different and the opencode models list has a bunch of invalid entries
github-actions commentedon Apr 7, 2026
To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.
sergiotapia commentedon Apr 20, 2026
Any idea when this will be fixed? I still can't use kimi 2.6 even though it's on my Zen subscription.
andrescevp commentedon May 3, 2026
up
magicprinc commentedon May 5, 2026
no deepseek on my Go subscription
1.14.34 (from choco)
opencode models --refreshmitjap commentedon Jun 17, 2026
Just a heads-up if there is anyone else wandering around. Model list is fetch form models.dev which might not be up to date with latest changes. You can manually edit the models.json and manually add models that you are missing. This file might be overwritten so make a backup to avoid duplicate work.
Gaploid commentedon Jul 11, 2026
Hey folks, I’m thinking we could quickly build a models.dev-compatible endpoint based on our data.
We automatically ingest model data primarily from LLM providers, with models.dev and LiteLLM JSON files as fallback sources. This could serve as a drop-in replacement for the existing endpoint, allowing new models to appear almost immediately after release with actual pricing. As a fallback opencode can rely on local file of models.dev.
What do you think?
We’ve already built something similar for LiteLLM: https://cloudprice.net/models/api?litellm=1
eezzekl commentedon Sep 21, 2026
NVIDIA NIM — mismo problema de cache stale
Confirmo que este bug afecta también a NVIDIA NIM (proveedor
nvidia/nimen models.dev).Síntoma: modelos retirados/actualizados en el catálogo de NVIDIA siguen apareciendo en
/modelsy en el picker de la TUI días después, hasta que se borra~/.cache/opencode/models.jsono se correopencode models --refresh.Impacto real: intentos de usar modelos deprecated (p.ej.
nemotron-3-ultraviejo) fallan en runtime con 404/400, rompiendo sesiones y agentes que los referencian.Workaround actual: alias
ocm='opencode models --refresh'en shell y recordatorio mental antes de empezar trabajo con NIM.Relacionados v2: #2047 (LM Studio ghost models), #7714 (provider cache no actualizable), #42376 (startup bloquea fetch models.dev), #35090 (models.dev sin retry/cache/fallback).
Pedido: exponer
opencode models --refreshcomo comando interno accesible desde la TUI (p.ej./model refresho keybind), y/o TTL configurable para proveedores externos de alta rotación como NIM.