Continue on one prepaid key

• The Continue extension in VS Code or a JetBrains IDE. Continue writes ~/.continue/config.yaml the first time it runs (%USERPROFILE%\.continue\config.yaml on Windows), per its configuration docs (read 2026-10-28).
• A RelaySwitch key. Sign in with an email code at the console and create one. Keys start with sk-relay-; one key is enough for every model in this guide.
• A balance. Top-ups start at $10, paid by card or in crypto, and the balance is spent per token as you go. No subscription, no KYC. Before the first top-up the console chat gives you 12 free turns a day to try models; requests from Continue bill the balance.

1. Open the Continue sidebar (Cmd/Ctrl+L in VS Code, Cmd/Ctrl+J in JetBrains), open the agent selector above the chat box and click the gear icon on the local agent. That opens config.yaml.
2. Replace the models list with the block below. Keep name, version and schema: v1 at the top; Continue requires all three.
3. Paste your key over sk-relay-... in every entry, or define it once as a YAML anchor and reuse it.
4. Save. Continue reloads the file on save; the three models appear in the model picker under the names you gave them.

Field names follow Continue's openai provider docs and its config.yaml reference (both read 2026-10-28). apiBase keeps the /v1, and Continue appends /chat/completions itself. The model value is the id exactly as the Models page shows it, not the display name. name is yours to choose. To add a model, copy an entry and change name and model.

The older config.json is deprecated in Continue; if yours still has one, the reference links a migration guide.

• roles: which jobs the entry is offered for. Continue's default is [chat, edit, apply, summarize], so you can leave the field out. Set it when one model should only chat and another should only apply edits.
• capabilities: tool_use lets the entry run in Agent mode, where Continue calls tools and edits files on its own. Continue tries to detect it; listing it removes the guess. Add image_input only for a model the table marks as reading images.
• defaultCompletionOptions: contextLength tells Continue how much context it may pack into a prompt, and maxTokens caps one answer. Billing ignores both; every request bills at the live catalog price.

Set contextLength no higher than the model's real window. A higher value lets a long Agent session grow past what the model accepts, and the request fails as too long.

Two curl calls rule out the key and the base URL before you debug the YAML. The first lists every model id your key can call:

The second sends one small request the way Continue does. -i prints the headers, including the request id and, on a non-streamed answer, the charge in USD.

If both work, anything left is in the file. Send a message in Continue and the request shows on the console's Usage page within seconds, with its tokens and its charge against the balance.

Continue resends the conversation, the files it pulled in and the tool results on every step, so input tokens are most of the bill. A low input price matters more than a low output price. Of the models here, GLM 5.3 has the lowest input rate today at $0.3178 per million tokens.

Claude ids answer the OpenAI format too, so Claude Sonnet and Opus work through provider: openai with no Anthropic entry. For the full list, with every price and context size, see the Models page. How the prepaid balance and the cached rate work is on Pricing.

The task: one feature built over 40 Agent turns, late enough in the session that the context is large. Each turn sends 80,000 tokens of earlier context, 2,000 new input tokens, and gets 1,500 output tokens back. Those sizes are an assumption; your own Usage page shows the real ones.

Two columns, because whether the earlier context hits the prompt cache depends on the model and the request. When it does, the response reports cached tokens and they bill at the cached rate. The console's Usage page shows the split for every request.

The spread is wide. With cache hits the same task runs $0.2904 on DeepSeek V4 Pro 0423 and $1.26 on Claude Opus 5.5. Keep a cheap entry for routine edits and a dear one for the hard step; switching is one click in the model picker, and both draw on the same balance.

• 401: the key is wrong, revoked or cut short. Paste it again from the console and run the curl check.
• 402: the balance is too low for the request. Top up in the console by card or crypto; the key works again at once.
• 404: the model value is not in the catalog. Check for a typo or a display name in place of the id.
• 429: the key hit its rate limit. Wait a minute and resend, or create a key with a higher Requests / min in the console.
• Connection errors: apiBase is missing /v1, has a typo, or a proxy or firewall blocks it. The curl check from the same machine tells you which.
• The model is missing from the picker: a YAML indentation error. Each entry starts with - name: and the fields below it line up two spaces in.
• A GPT entry misbehaves: Continue sends gpt-5 models to /responses by default, which RelaySwitch serves. To use /chat/completions instead, add useResponsesApi: false to that entry.
• The request is refused as too long: contextLength is set higher than the model accepts. Use the number from the table.