Cursor Model List Out of Date vs Official Site

Q: I clicked "Add Models" and it's still not there. Now what?

That rules out the cache. Check your plan against [cursor.com/docs/models-and-pricing](https://cursor.com/docs/models-and-pricing) (Step 3), then look for an active rollout or incident on [forum.cursor.com](https://forum.cursor.com) and the [status page](https://status.cursor.com) (Step 4). If others on your exact plan already have it, clear the cache directories and relaunch.

Cursor's site advertises a model but your local Settings → Models picker doesn't show it. Old client, stale cache, or staged rollout — here's how to tell which and fix it.

Published: May 21, 2026 Updated: Jun 21, 2026 Author: AI Productivity Guide Team 🌐 查看中文版本

The Cursor changelog headline says it now supports a new model (say Claude Opus 4.7 or a fresh gpt-5-codex variant). You open the model picker in Composer and it isn’t there, or the list still shows older entries only. Your account isn’t broken. Nine times out of ten it’s one of three things: the desktop client is behind, the cached model manifest is stale, or the model is in a staged rollout that hasn’t reached your plan/region yet.

Fastest fix (works for most people): update to the latest Cursor build (3.8 or newer as of June 2026), fully quit and relaunch, then open Cursor Settings → Models and click Add Models to force a fresh fetch. If the model still isn’t listed, it’s a rollout or plan issue — jump to Step 3.

Why “Add Models” and not just the dropdown: the in-chat picker shows only models you’ve enabled. The full, server-fetched catalog lives behind the Add Models button in Cursor Settings → Models. A model can exist in the catalog but be hidden from the dropdown until you add it.

Common causes

1. Cursor client too old

New model support needs matching client code (routing, params, UI labels). An old client has no idea the model exists. Auto-update silently fails more often than you’d think, and enterprise installs frequently pin versions.

How to judge: open Cursor → About Cursor for your version and compare with the changelog. As of June 2026 the latest is 3.8 (shipped June 18, 2026). More than 2 minor versions behind, or still on the 2.x line, is old.

2. Model list cache hasn’t refreshed

Settings → Models pulls a model manifest from Cursor’s backend and caches it locally. Even right after an update, the cache can keep serving a stale list until you force a refresh with Add Models (or restart).

How to judge: a post-update relaunch still shows the old list, and signing out and back in doesn’t help. That points at the cache rather than your version or plan.

3. Staged rollout by region or plan

New models usually land on higher tiers first (Pro / Pro+ / Ultra / Business) and on US infrastructure before other regions, then widen out over days. During that window you genuinely don’t have access yet, no matter how many times you refresh.

How to judge: search forum.cursor.com for <model name> rollout to see if others are waiting too. The status page does not list rollouts as a permanent component, but it does log recent model incidents (for example, June 2026 entries for Composer 2.5 degradation and “Elevated errors across many models on Anthropic”), so check it for active model problems.

4. Plan doesn’t include the model

Some models or modes are gated to higher tiers — Max mode on the heaviest Anthropic/OpenAI models, and certain reasoning models, can be Business / Enterprise only, so a Hobby or Pro account won’t see them. A few are reachable only through your own API key (BYOK).

How to judge: check Cursor Settings → Account for your plan, then cross-reference the official Models & Pricing page. Note: that page lists every model with its provider and token cost but no longer prints a per-plan “Available on” column, so confirm gating in-app and on the pricing page rather than expecting a tidy matrix.

5. BYOK filters the list down

When you supply your own provider key, the picker shows only what that key actually exposes. Cursor’s in-house models (the Composer line) and Cursor-routed-only entries won’t appear under a raw provider key. BYOK also unlocks chat completions for standard models only — Agent and Edit rely on Cursor-routed custom models, and Tab completion cannot run through your key at all.

How to judge: in Cursor Settings → Models, each entry indicates whether it runs through Cursor or your own key. If the in-house Composer models vanished after you added a key, BYOK filtering is the cause.

6. Network restriction / VPN

Some regions and corporate networks block Cursor’s backend, so the client can’t fetch the manifest and falls back to a minimal built-in list.

How to judge: open View → Output, pick Cursor in the dropdown, and look for fetch failures (for example, “failed to fetch models” or connection-timeout lines). Switch to a different network or disable the VPN and retry.

Which bucket are you in

Match your symptom to the most likely cause before touching anything:

Symptom	Most likely cause	Go to
You’re on the 2.x line or several versions behind	Client too old	Step 1
Updated and relaunched, list still stale	Cache not refreshed	Step 2
Model exists in the catalog but not in the dropdown	Not yet added	Step 2 (Add Models)
Coworkers on the same plan see it, you don’t	Cache / region rollout	Step 2, then Step 4
Nobody on your plan sees it; higher tiers do	Plan gating	Step 3
In-house Composer models disappeared after adding a key	BYOK filtering	Step 5
`View → Output → Cursor` shows fetch failures	Network / VPN block	Step 6

Before you start

Confirm whether it’s only your account or everyone on your team. Widespread = rollout or plan, not a local fix.
Commit settings.json before updating so you can tell what a new default changed.
Note your current Cursor version, plan, and network (home / corporate / VPN).

Info to collect

Cursor version (Cursor → About Cursor) and plan (Cursor Settings → Account).
The exact target model name as it appears on cursor.com/docs/models-and-pricing.
A screenshot of your current Settings → Models list (including the Add Models picker).
The last ~50 lines of View → Output → Cursor (look for fetch errors).
Your BYOK status and whether the provider’s own console shows the model.

Shortest fix path

Work the steps in order: update, refresh the catalog, then rule out plan and rollout.

Step 1: Update to the latest build

Cursor → Check for Updates (it’s under the application menu on macOS; the Help menu on Windows / Linux). Update and relaunch.

If it claims “you’re up to date” but the changelog shows a newer release, the in-app updater is stuck. Reinstall over the top:

# macOS (Homebrew install)
brew upgrade --cask cursor

# Or download a fresh installer from https://cursor.com/download

On Windows, download the installer from the same page and run it over your existing install. On Linux, replace the AppImage with the latest from the download page. For pinned enterprise installs, the version is controlled centrally — ask IT to bump it.

Confirm via Cursor → About Cursor that you’re now on 3.8 or newer (as of June 2026).

Step 2: Refresh the model catalog

The dropdown only shows models you’ve enabled, and the catalog is cached. Force a fresh fetch:

Option A (preferred): Cursor Settings → Models → Add Models
  → this re-fetches the full server catalog; enable the model you want.
Option B: sign out and sign back in (bottom-left avatar → Sign out, then Sign in).
Option C: fully quit Cursor (Cmd/Ctrl+Q, not just close the window) and relaunch.

If those don’t refresh the list, clear the cache directories and relaunch:

# macOS
rm -rf ~/Library/Application\ Support/Cursor/Cache
rm -rf ~/Library/Application\ Support/Cursor/Code\ Cache

# Windows (PowerShell)
# Remove-Item -Recurse -Force "$env:APPDATA\Cursor\Cache","$env:APPDATA\Cursor\Code Cache"

# Linux
# rm -rf ~/.config/Cursor/Cache ~/.config/Cursor/"Code Cache"

These only clear caches, not your settings or sign-in. Relaunch afterward.

Step 3: Confirm your plan actually includes the model

Open cursor.com/docs/models-and-pricing and find your target model. The page lists provider and token cost but no longer prints a per-plan availability column, so confirm tier gating two ways:

In-app: Cursor Settings → Account shows your current plan (Hobby / Pro / Pro+ / Ultra / Business).
On the web: the pricing page describes what each tier unlocks.

Heavy models in Max mode and some reasoning models can be limited to higher tiers, so a Hobby or Pro account may simply not be entitled to them. If the model is above your tier, the list is behaving correctly — upgrade the plan or use BYOK (Step 5).

Step 4: Check for an active rollout or incident

If you’re up to date, the catalog is refreshed, and your plan qualifies but the model still isn’t there, it’s likely mid-rollout. Search forum.cursor.com for <model name> rollout to see whether others are waiting. Check the status page for active model incidents — it logs things like Composer 2.5 degradation and provider-wide Anthropic errors even though it has no permanent “rollout” component. If it’s a rollout, waiting a day or two is the fix; if it’s a logged incident, wait for it to clear.

Step 5: Use BYOK as a bridge

To get a model before its Cursor-routed rollout reaches you, add your own provider key:

Cursor Settings → Models → API Keys → paste your Anthropic / OpenAI key → Verify
→ enable the desired model entries

Know the limits before you rely on this:

BYOK unlocks chat completions for standard models only; the in-house Composer models won’t appear under a raw provider key.
Agent and Edit depend on Cursor-routed custom models, so some agentic features can degrade on BYOK.
Tab completion never routes through your key.

Step 6: Rule out a network / VPN block

If View → Output → Cursor shows fetch failures (for example, “failed to fetch models” or connection timeouts), the client can’t reach the catalog endpoint:

Disable the VPN, or switch to a different node, and relaunch.
A corporate proxy may block Cursor’s backend. Set the proxy in Cursor Settings (search “proxy”), or have IT allow Cursor’s domains (cursor.com, cursor.sh).

How to confirm it’s fixed

Cursor → About Cursor shows the latest version (3.8 or newer as of June 2026).
Cursor Settings → Models lists the target model after Add Models (enabled, not greyed out).
The in-chat model picker in Composer shows it as selectable.
Send a one-line prompt against that model. A normal reply (no “model not available” or “this model is not enabled”) confirms end-to-end routing works.

If it still fails

Reduce the repro to a single prompt against a single model so you can describe it cleanly.
Roll back to the previous Cursor build to check whether a new release quietly dropped the model.
Search forum.cursor.com for <model name> not showing; include your Cursor version, plan, and region.
Capture View → Output → Cursor logs and post them to Bug Reports — the models team monitors that category.

Prevention

Check for updates roughly monthly; don’t rely on auto-update alone, since it can silently fail.
Follow the changelog and forum.cursor.com so you know when a model ships and how rollouts are pacing.
When a model isn’t available yet, use BYOK as a bridge until the Cursor-routed path opens.
For teams, make sure the plan tier covers the models people actually use before they need them.
Allow Cursor’s backend domains (cursor.com, cursor.sh) in corporate proxies.

FAQ

Why does the model show on cursor.com but not in my editor? The website and changelog announce a model the moment it ships to the platform. Your editor only sees it once your client is current, the cached catalog has refreshed (click Add Models), and the rollout has reached your plan and region. Those three lag behind the announcement by anywhere from minutes to a few days.

I clicked “Add Models” and it’s still not there. Now what? That rules out the cache. Check your plan against cursor.com/docs/models-and-pricing (Step 3), then look for an active rollout or incident on forum.cursor.com and the status page (Step 4). If others on your exact plan already have it, clear the cache directories and relaunch.

Does signing out and back in refresh the model list? Sometimes. It re-pulls your account state and can refresh the catalog, but it’s less reliable than Add Models or clearing the cache. Try it, but don’t stop there if the list is still stale.

Will BYOK let me use the in-house Composer models? No. A raw provider key only exposes that provider’s standard chat models. Cursor’s in-house Composer line (Composer 2.5 as of June 2026) runs through Cursor’s own routing, so it won’t appear under your own key, and Tab completion can’t use your key either.

The model is in my list but greyed out — is that the same bug? No. Greyed-out usually means it’s enabled but gated to a higher plan or mode (for example, Max mode), or temporarily unavailable due to an incident. Check your plan (Step 3) and the status page rather than refreshing the catalog.

Tags: #Cursor #Troubleshooting