Cursor’s Models settings accept a custom OpenAI API key and an overridden base URL. Set the key to a Continuum API key (they are prefixed cont_sk_ and created in the Continuum web app under Settings, then Account, then Inference API) and the base URL to https://continuumcode.ai/v1. Cursor checks the pair against GET /v1/models when you save it, which the Continuum gateway serves. Add each model ID you want by hand under Models, Add model, because Cursor only pre-lists OpenAI’s own names. Chat and agent then run on Continuum, including Claude models, which go through the OpenAI slot rather than Cursor’s Anthropic field. Cursor Tab and Cursor’s own Composer models are unaffected; they stay on your Cursor plan. Personal Continuum API keys are a paid-plan feature: Plus is $25/mo, Max 100 is $100/mo, Max 200 is $200/mo, and Ultra is $500/mo as of August 2026.
- Two fields do the whole job: the OpenAI API Key, and Override OpenAI Base URL set to
https://continuumcode.ai/v1. - Continuum keys are prefixed
cont_sk_and the full key is shown exactly once, at creation. - Everything goes through the OpenAI slot, Claude models included. Leave Cursor’s Anthropic key field empty.
- Cursor pre-lists only OpenAI names, so type each model ID by hand and pin one rather than leaving it on Auto.
- Tab and Cursor’s own models are untouched. You are replacing the third-party model pool, not your Cursor plan.
- Responses are capped at 8,192 output tokens. Agent turns iterate, so in practice only long single generations notice.
What the override actually changes
Cursor splits its models into two pools. Its own models, Composer and Cursor Grok, come with a paid Cursor plan. Everything else, the Claude and GPT and Gemini families, is the Other Models pool, drawn from a dollar allowance and then billed on demand. The base URL override redirects that second pool, and nothing else.
What runs where once the override is on, checked 13 August 2026.
| Surface | Where it runs | Who bills it |
|---|---|---|
| Chat and Agent, on a model you added | Continuum’s inference API | Your Continuum plan |
| Tab | Cursor, unchanged | Your Cursor plan |
| Composer and Cursor Grok | Cursor, unchanged | Your Cursor plan |
| Auto | Cursor’s own routing | Your Cursor plan |
Nothing is installed locally for this. There is no CLI, no proxy on a port, and no daemon: Cursor talks to https://continuumcode.ai/v1 directly, the same way it would talk to OpenAI.
What you need first
Setting Cursor up this way requires Continuum Plus or higher. Inference keys do not mint on the Free tier, so the Create key button stays disabled until you subscribe, and everything below starts with a key.
- A Continuum account on a paid plan. Personal API keys are a paid-plan feature. As of August 2026 that is Plus at $25/mo, Max 100 at $100/mo, Max 200 at $200/mo, or Ultra at $500/mo; each step up raises the weekly hosted allowance. See pricing for what each one includes.
- Cursor installed and signed in. Any recent version; the fields used here live in Settings under Models.
- Nothing else. No local install, no Continuum app required, no network configuration.
Point Cursor at Continuum
Create a Continuum API key
In the Continuum web app at continuumcode.ai/app, open Settings, then Account, then Inference API. Create your key under Your keys. The same panel is on the Usage tab. Copy it immediately: it begins with cont_sk_, and the full value is displayed exactly once. If you lose it, you delete that key and make another; there is no way to read it back.
Keys mint on Plus and above, so on the Free tier Create key stays disabled until you subscribe.
Open Cursor’s model settings and paste the key
In Cursor, click the gear icon to open Settings, choose Models, and scroll to the API Keys block. Paste the cont_sk_ key into the OpenAI API Key field. This is the correct field even if the only thing you plan to run is Claude.
Turn on the base URL override
Enable Override OpenAI Base URL and enter https://continuumcode.ai/v1 exactly, with no trailing slash and no path after /v1. Cursor appends the rest of the route itself.
Check the pair yourself
Cursor gives you no explicit verify control for a custom endpoint, so run the check it cannot show you — one command in any terminal:
curl -s https://continuumcode.ai/v1/models \
-H "Authorization: Bearer cont_sk_..."A JSON list of model IDs proves three things at once: the key is valid, the URL is right, and your plan is entitled to the hosted models. An error here is worth fixing before you touch Cursor again, because from inside the editor every later problem looks the same.
Add the model IDs you want
Still under Models, use Add model and type each ID exactly as listed in the next section. Cursor only pre-lists OpenAI’s own model names, so most of these appear only after you add them. One wrinkle: an ID Cursor already lists — the GPT families, mostly — cannot be re-added, and Cursor tells you the model is already saved. That is fine: enable the built-in entry instead, and with the override on it routes to Continuum like any other. Beyond those, disable Cursor’s remaining built-in entries: a name Continuum does not serve fails only at request time, which is a confusing way to find out.
Pick the model and work normally
Choose one of your added models in chat or agent mode. Pin it rather than leaving the selector on Auto, since Auto is Cursor’s own routing and will not choose a model it did not put there. Tool calling works, so agent mode behaves as it always did.
The models to add
These IDs were read from the live gateway on 17 August 2026. Type them exactly; a near-miss is the single most common failure after setup, and it surfaces as a request-time error rather than a picker that refuses the name.
Paid lane, drawn from your plan’s inference allowance.
| Model ID | What it is |
|---|---|
claude-opus-5 | The flagship, and the recommended default |
claude-fable-5 | Claude 5 family |
claude-sonnet-5 | Claude 5 family |
gpt-5.6-sol | GPT-5.6 family |
gpt-5.6-terra | GPT-5.6 family |
gpt-5.6-luna | GPT-5.6 family |
grok-4.6 | xAI |
gemini-3.5-flash | |
deepseek-v4-flash | DeepSeek |
glm-5.3 | Z.ai, via OpenCode Go |
muse-spark-1.2 | Muse |
The free lane
These models do not draw down your plan’s weekly inference allowance. Reaching them from Cursor still needs a paid key, because that is what minting a key requires. They are chat-grade rather than agent-grade: useful for questions, explanations, and quick edits, and not what you want driving a long agent run.
On a paid key, these :free models do not draw your weekly budget.
| Model ID | Provider |
|---|---|
google/gemma-4-26b-a4b-it:free | |
nvidia/nemotron-3-ultra-550b-a55b:free | NVIDIA |
nvidia/nemotron-3-super-120b-a12b:free | NVIDIA |
cohere/north-mini-code:free | Cohere |
poolside/laguna-m.1:free | Poolside |
tencent/hy3:free | Tencent |
curl -s https://continuumcode.ai/v1/models \
-H "Authorization: Bearer cont_sk_..."What Continuum meters, and what Cursor cannot see
This setup leaves you with two meters that do not know about each other, which is worth understanding before it surprises you mid-month.
| Question | Where the answer lives |
|---|---|
| How much of my Continuum allowance is left? | The Usage tab in the Continuum web app |
| How much of my Cursor allowance is left? | Cursor’s own usage dashboard, which now only counts Tab and Cursor models |
| What did that request cost? | Continuum’s side. Cursor’s per-request display does not know Continuum’s rates |
Two limits are worth stating plainly. Output is capped at 8,192 tokens per response, so a single generation asked to emit an entire file can truncate; agent turns iterate rather than emitting one enormous response, so they rarely meet the ceiling. And this is not a way to stop paying Cursor: Tab is the feature most people actually stay for, and Tab does not route through Continuum.
When it does not work
| Symptom | Cause | Fix |
|---|---|---|
| The models curl fails to connect | Typo in the base URL, or a trailing slash | Set it to exactly https://continuumcode.ai/v1 |
| The models curl returns an auth error | Key mistyped, revoked, or truncated on paste | Mint a new key; the original is unrecoverable |
| The models curl works, Cursor requests fail | Plan or entitlement gating, or a model-name mismatch | Re-run the curl: the models endpoint re-runs the same gate and its list is authoritative |
| Unknown or invalid model | The ID does not match what the gateway serves | Check it against GET /v1/models |
| A model you did not add fails | A Cursor built-in entry is still enabled | Disable Cursor’s built-in models while the override is on |
| A long answer stops mid-sentence | The 8,192-token output cap | Ask for the work in parts, or use agent mode |
| Claude models do nothing | The key went in the Anthropic field | Clear it; every model goes through the OpenAI slot |
| The override connects, but requests fail or the stream dies mid-response | Cursor’s HTTP/2 transport against a custom base URL | Cursor Settings, then Network, then HTTP Compatibility Mode: set it to HTTP/1.1 |
| You want Cursor’s own models back | The override is global while it is on | Turn Override OpenAI Base URL off; one toggle restores stock behaviour |
Everything else about Cursor keeps working. Your Cursor account, Tab, Composer, and the Cursor plan features are untouched, because the override redirects the Other Models pool’s API traffic and nothing more. Turning the override off restores Cursor’s stock behaviour with one toggle.
Questions people ask
In the Continuum web app at continuumcode.ai/app, open Settings, then Account, then Inference API, and create your key under Your keys. The same panel is on the Usage tab. Keys are prefixed cont_sk_ and the full value is shown once at creation, so copy it before closing the dialog. Create key is disabled on the Free tier: minting a key needs Plus or higher.
The OpenAI API Key field, together with Override OpenAI Base URL set to https://continuumcode.ai/v1. Everything is served through the OpenAI-compatible surface, including Claude models, so nothing goes in Cursor’s Anthropic key field.
Yes. Tool calling is supported, so agent mode behaves as usual. The gateway serves POST /v1/chat/completions and POST /v1/responses, which is what Cursor uses.
That pattern almost always means plan or entitlement gating rather than a bad key or URL. The models endpoint re-runs the same gate, so re-running the curl surfaces the real error instead of the generic request-time failure.
No. The override only redirects the third-party model pool. Cursor Tab and Cursor’s own Composer models still come from your Cursor plan, and Tab is the feature most people are actually paying for.
Run curl -s https://continuumcode.ai/v1/models with your key as a bearer token. Trust that over any static list, including the one in this guide, because model lineups change faster than documentation.
Responses are capped at 8,192 output tokens. Long single generations can truncate at that ceiling; agent turns iterate across several responses, so in practice they are unaffected.
They are free of your allowance, not free of a plan. On a paid key the models tagged :free do not draw your weekly budget, but reaching them from Cursor still needs a key, and keys mint on Plus or higher. They are also chat-grade rather than agent-grade, so they suit questions and small edits rather than long agent runs.
Not accurately. Cursor prices requests against rates it knows, and it has no way to learn a third-party endpoint’s. Usage is metered on the Continuum side and visible on the Usage tab.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Cursor docs: models the model pools and the custom API key fields
- Continuum pricing plan names, prices, and which tiers include personal API keys
- Continuum hosted inference the model lanes and the gateway surfaces, read 13 August 2026; the live /v1/models endpoint itself needs a cont_sk_ bearer