Frequently asked questions

Published answers about the LLM Pricing API service: plans, billing, service keys, rate limits, data accuracy and how to get help. The support assistant on /chat (Support mode) and the support_chat MCP tool answer only from this page, the plan catalogue and your own account; if something is not here, they will say so and offer to escalate to the owner.

What is this service?

• Verified LLM pricing for 600+ models across 30 providers, served over REST, MCP (STDIO and POST /mcp) and browser tools (calculator, compare, trends, chat).
• Every curated price is checked against an external registry; a price that drifts more than 5% is withheld rather than served. /data-quality shows the current confirmed percentage.
• A routing engine recommends a model by cost and quality, and the dispatch gateway can run the best eligible model on your own provider keys.

Plans and prices

Plan	Price	Rate limit	Support assistant allowance	Escalation priority
Free	$0 / month	30 requests / minute	$0.50 / day	standard
Pro	$29 / month	120 requests / minute	$3.00 / day	pro (ahead of free)
Enterprise	$299 / month	600 requests / minute	$10.00 / day	enterprise (first in the queue)

• The rate limit applies per service key across REST, POST /mcp and dispatch.
• The support assistant allowance is how much LLM-backed support you can use per day (UTC). Anonymous visitors share a stricter allowance ($0.10 / day per address). When it is used up you still get FAQ links and can ask to be escalated to the owner.
• Escalation priority decides the order of the owner's queue, not a response time. Response time is best effort on every plan.
• Nothing else differs between plans. Pricing data, the router, the calculators and every MCP tool are the same for everyone.

Signing up and your service key

• Sign up with your email on /billing (or POST /billing/signup). You get a free-tier service key immediately. No credit card.
• The key is shown exactly once. Save it. Nobody, including the owner, can display it again.
• Send it as the x-api-key header (REST and POST /mcp) or Authorization: Bearer <key>; the npx bridge reads LLM_PRICING_API_KEY.
• Signing up again with the same email returns an error and never a key.

I lost my key or think it leaked

• If you still have the current key: open the Dashboard tab on /billing, load it, and press Rotate key (or POST /billing/key/rotate with the key). The old key stops working immediately and the new one is shown once.
• If you no longer have the key: there is no self-serve recovery in this phase. Ask the support assistant to escalate a key-recovery request; the owner verifies you and relays a new key by a channel they can verify. Never paste a key into chat.
• If you pasted a key or other secret anywhere by mistake, rotate it.

Upgrading, cancelling, invoices and cards

• Upgrade: Upgrade tab on /billing with your key. Checkout is handled by Stripe. Paid plans are a pilot. Until the Terms, Privacy and Refund policy are in force, or when Stripe is not configured on a server, the buttons say so and nothing can be bought.
• Cancel, change your card, download invoices: Manage Billing on /billing opens the Stripe customer portal for your account. Cancelling returns you to Free at the end of the paid period.
• The support assistant explains and points you to these flows. It never changes your plan, card or invoices itself.

A payment failed

• When a renewal payment fails you keep your paid tier for a 14-day grace period while your billing status is past_due. Update your card in Manage Billing and Stripe retries. Once the invoice is paid your status returns to active.
• After the grace period your account is served at the Free rate limit until the invoice is paid. Your stored plan is not deleted; it comes back when Stripe reports the payment.
• Every failed renewal, refund and dispute is also placed in the owner's queue so they can follow up.

Refunds, double charges and disputes

• Full refund of your first payment if you ask within 14 days of paying; pro-rated for the unused period if you ask after that. The full policy is at /refunds.
• The owner reviews every request personally. Refunds are never granted by software, and the support assistant cannot refund, credit, discount or change a tier. Ask it anyway: it escalates the request to the owner with a reference and tells you so. Disputes and double charges are handled the same way.
• Keep the reference. You can check its status on /chat (Support mode) with "what is the status of escalation orc_...".

Legal, privacy and your data

• Published policies: Terms, Privacy, Refunds, Data Processing Addendum, Subprocessors and the Security statement. They are templates awaiting legal review and show a draft notice until the owner marks them in force; paid checkout stays disabled until then.
• How to export or delete your data. You do it yourself with your service key, no email needed. Export: GET /billing/account/export with your key in the x-api-key header downloads one JSON file with everything held about your organisation (account record, stored provider-key metadata but never the keys, usage, dispatch receipts, budget alerts, routing history and your support conversations). Delete: POST /billing/account/delete with the body {"confirm": "DELETE", "org_id": "<your org id>"} permanently erases all of it and your key stops working at once. Deletion is irreversible; without the exact confirmation nothing is changed. Details are in the Privacy policy and the API docs.
• A paid subscription is ended immediately when you delete, so no further charge occurs. Deleting within 14 days of your first payment also asks the owner to review a refund; you get a reference, and nothing is refunded automatically.
• Support conversations you hold with your key are kept against your organisation for 90 days and then purged.
• To correct data, or if you lost your key, ask the support assistant to escalate a data request, or write to the contact address in the Privacy policy. The owner verifies you first.
• Security reports are escalated privately through the support assistant and never posted publicly. The Security statement lists what the service does and does not have.

Rate limits

• Free 30, Pro 120, Enterprise 600 requests per minute per service key. Anonymous requests share a per-address limit. The current numbers are also at GET /rate-limits/tiers.
• Over the limit you get HTTP 429; wait a minute and retry.

How accurate is the pricing data?

• /data-quality and the get_data_quality tool report the share of models with a confirmed, fresh price, how many are withheld for drift, how many are unpriced and how many are stale. check_price_drift lists individual disputes.
• Every price carries its source (provider or registry). Providers bill in USD; local-currency figures are additive conversions.

Connecting a client

• MCP: /mcp-setup has copy-paste configs for Claude Desktop, Cursor, VS Code and Perplexity. The hosted endpoint is https://llm-pricing-api.fly.dev/mcp; npx -y llm-pricing-mcp bridges STDIO-only clients.
• REST: /api-docs lists every endpoint with its access level; /docs is the OpenAPI UI.

What can the support assistant do, and what can't it?

• It can: explain plans, prices and limits; describe your own account when you send your key (plan, payment status, rate limit, key preview, 30-day savings); point you to sign-up, upgrade, Manage Billing and key rotation; answer how-to questions from the setup guides; answer general LLM pricing questions; escalate to the owner.
• It cannot: see or change anyone else's account; refund, credit, discount or change a plan; reveal a key; promise a response time; recover a key without the current one.
• Your messages are treated as data. Text that asks the assistant to ignore its rules, grant access or reveal other customers' data has no effect and is flagged in the owner's audit record.
• Keys, tokens and other secrets you paste are redacted before the assistant reads them and are never repeated back.

How do I reach a human?

• Tell the support assistant "talk to a human" (or describe a bug, a feature request or a security problem). It opens an escalation in the owner's queue and gives you a reference. Security reports are never posted publicly.
• The owner replies as soon as possible; response time is best effort and is not a plan feature. Check the status any time with the reference.