Last verified

Cline Bot Inc. · Pricing

Cline pricing: the agent is free, and the cheap model has terms

Cline itself is never sold — Apache-2.0, no paid edition — so every route below buys model access. Seven of them exist: your own hardware, your own key, the vendor's dollar-metered gateway, its flat monthly pass, a Claude or an OpenAI subscription you may hold already, and an enterprise seat. Its published rate card for other vendors' models matched ours on twelve rows of fourteen.

§ 00 / YOUR CASE

What it costs you.

Full calculator →

Opens on a coding-agent day rather than a chat day: an agent re-sends the whole conversation every turn, so the input side is far larger than a chat workload and mostly cache reads. Cline Bot Inc. publishes no per-developer spend figure, so the preset is calibrated against the only one anybody publishes — Anthropic's. Change them to your volume; every plan re-judges as you type.

InputsUSD · US web-plan prices
Ecosystem

API side uses the vendor's published per-token list prices with prompt caching modeled (chat apps resend conversation history every turn — set cache to 0 for a stack without caching). Thinking models bill hidden reasoning tokens at the output rate — that's real spend in both worlds.

VerdictClaude · 80 msgs/day
Same usage at API rates
$203/mo
Claude Opus 4.8 · 2.4K requests/mo

Best value: Claude Max 5x — this volume would bill ×2.0 its price at API rates.

  • Claude Pro$20.00/mo break-even 8/day plan saves $183/mo
  • Claude Max 5x$100/mo break-even 39/day plan saves $103/mo
  • Claude Max 20x$200/mo break-even 79/day ≈ break-even

Plans cap usage in rolling windows and vendors don't publish those caps in tokens — verdicts compare the price of this volume, not entitlements. The higher the multiple, the more likely a cap cuts in first.

§ 01 / EVERY WAY TO BUY IT

Seven ways in, and the tool is on sale in none of them.

Every row buys a model, not the agent. Two are the same shape for different reasons: one is your key at your vendor, the other is the vendor's own gateway, which charges per request rather than per month and reports the cost of each call in dollars. Notice what the table cannot show — the two cheapest rows are also the two where the vendor's terms take the most, which is the subject of the last section on this page.

Route Bought as Price What it buys
Local model Own hardware Your hardware The only route that needs no account. The vendor sizes it in memory rather than money: 16-32GB of RAM for small or quantized models, 32-64GB for mid-size coding models, 64GB and up for larger models and bigger context windows. It also tells you to switch on Compact Prompt, which is an admission that the agent's normal prompt is too big for this route.
Your own API key API key At model rates Anthropic, OpenAI, Google, AWS Bedrock, GCP Vertex, Azure, Groq, Cerebras, DeepSeek, OpenRouter or any OpenAI-compatible endpoint. This is also the only route the vendor's own terms describe unconditionally: bring your own key and route the calls through infrastructure you control, and Cline states it does not receive or store your tokens or your code.
Cline gateway (pay as you go) API key At model rates Sign in once, top up a balance, and pick from the gateway's catalogue instead of managing keys. There is no monthly fee and no minimum: the API returns the cost of each request in dollars to six decimal places. The vendor's pricing page says inference is sold at cost, and it publishes no rate card of its own to check that against.
ClinePass Subscription $9.99/mo A flat monthly subscription to a fixed shortlist of open-weight coding models, metered against three windows at once and quantified only as a multiple: two to five times the usage the same money would buy at standard API rates. The page carries a second, unquantified charge as well - a processing fee that may apply on top.
Claude Pro Subscription $20/mo $17/mo billed annually Cline drives the Claude CLI on your behalf, so the subscription you already pay for covers the model and the agent is free on top. Max works the same way with a larger allowance. Nothing about the plan changes: the limits, the windows and the model list stay Anthropic's.
ChatGPT Plus Subscription $20/mo The same trick against OpenAI: sign in with the account behind your ChatGPT subscription and Cline routes through Codex without a key. The allowance is the subscription's, which means it is metered in OpenAI's credits and not in anything Cline controls.
Cline Enterprise Per seat Contact Sales SSO, org-wide provider configuration, MCP allowlists, audit logs, OpenTelemetry export and VPC or air-gapped deployment. The pricing page prints Custom and offers a form. It is also the only tier the terms exempt from the licence Cline otherwise takes over your content - and that exemption is written to apply to a per-seat subscription, which is this tier and not the $9.99 one.

Free three ways, and the third has a price that is not money. The client is Apache-2.0 forever. A local model costs nothing and signs in to nothing. The vendor also rotates limited-time free models for anyone with an account - but the quota is described only as limited, published in no unit at all, the free models do not work through the API, and the docs say usage of them may be used to help improve model performance.

Read off, in order of appearance: docs.cline.bot/overview, 2026-09-26 cline.bot/tos, 2026-09-26 docs.cline.bot/cline-provider, 2026-09-26 docs.cline.bot/clinepass, 2026-09-26 docs.cline.bot/anthropic, 2026-09-26 docs.cline.bot/openai, 2026-09-26 cline.bot/pricing, 2026-09-26 docs.cline.bot/free-models, 2026-09-26

§ 02 / WHICH ROUTE IS YOURS

Priced by who you are.

Trying it out without creating an account
Local model
You want to know whether the plan-then-act loop suits you before signing into anything, and you have a machine with memory to spare.
Your hardware

This is the genuine zero, and it is also the only route the vendor's terms leave entirely alone. The sizing is published in memory rather than money: small or quantized models at sixteen to thirty-two gigabytes of RAM, mid-size coding models at thirty-two to sixty-four, larger models and bigger context windows above that. One instruction in the setup guide is worth reading as a warning rather than a tip — switch on Compact Prompt — because it concedes that the agent's normal prompt is larger than this route comfortably carries.

A full working day on the model you already trust
Your own API key
You have an API key, a model you have settled on, and no interest in a second bill or a second account.
At model rates

The list of supported providers is long enough that this is a real choice rather than a formality, and the cost is entirely the model's: no platform fee, no per-token surcharge, nothing the agent adds. It is also the one arrangement the terms exempt unconditionally from the content licence, in a sentence naming both halves — your own key, and calls routed through infrastructure you control. Pay attention to the second half: using your own key through the vendor's gateway is a different arrangement from this one.

Many models, one balance, no subscription
Cline gateway (pay as you go)
You switch models constantly — a cheap one for edits, a frontier one for the hard refactor — and managing six API keys is the part you want to stop doing.
At model rates

You top up a balance and the agent bills against it per request. There is no monthly fee and no minimum, and the API reports each call's cost in dollars to six decimal places, which makes this the rare metered product whose unit is not a private token. What the vendor does not publish is the rate card: four hundred and fifty-eight model names are served on an open endpoint with no prices attached, and every pricing endpoint we tried needs a key. The pricing page's claim is that inference is sold at cost, and the nearest thing to evidence is the rate card on the pass, which stood up on twelve of fourteen rows.

Heavy use of open-weight models on a fixed budget
ClinePass
You are happy on GLM, Kimi, DeepSeek, MiniMax, MiMo or Qwen, you code most days, and you want one predictable line on the card.
$9.99/mo

The cheapest predictable route, and the one that needs the most reading. The allowance is quantified only as a multiple — two to five times the usage the same money buys at standard API rates — and it is metered against three windows simultaneously, a five-hour rolling one, a calendar week and a calendar month, none of which is visible outside the dashboard. Two more things belong in the decision. The page carries a second charge with no number on it, a processing fee that may apply. And the list of included models is published twice, in two versions that agree on nine of fourteen.

The subscription you are already paying for
Claude Pro
You hold a Claude subscription for chat, and you would rather not open a second inference account at all.
$20/mo
$17/mo billed annually

Cline drives the Claude CLI on your behalf, so the model is already paid for and the agent adds nothing. Max works identically with a larger allowance, and the same trick exists against OpenAI through Codex sign-in. The honest limit is that none of the metering is Cline's: the windows, the ceiling and the model list stay the subscription's, so this route cannot be tuned, only used. If the allowance is the binding constraint, the answer is a bigger plan or a key, not a different agent.

A team that needs the audit trail
Cline Enterprise
Several developers, a platform team that has to answer for where the code goes, and a procurement process that needs a contract.
Contact Sales

No number is published; the page prints Custom and offers a form. Two things make this tier structurally different rather than merely larger. It is the only one where the deployment can sit in your own VPC, on-premises or air-gapped, so the inference request never passes through the vendor at all. And it is the only one the terms exempt from the content licence — an exemption written for a per-seat subscription, which this is and the flat monthly pass is not. Note also that there is nothing between this and free: the pricing page has two columns and no third.

§ 03 / BRING YOUR OWN KEY

Bring your own key — and read what the cheap row is made of.

Cline is provider-agnostic by design, and the vendor says so more loudly than most: its enterprise page argues that most tools force you to buy inference through them with markup, and that this one connects directly to your providers with none. The claim is attractive and, as published, unverifiable — the gateway lists four hundred and fifty-eight model names on an open endpoint and no prices, and the endpoints that would carry rates require a key. So the question this page can actually close is the neighbouring one: when the vendor prints other companies' prices, is it printing them correctly?

It prints fourteen of them. The monthly pass publishes a reference table of per-million-token rates for the models it includes, describing them as the underlying rates and as an aid to understanding how usage is measured against the quota. We joined that table to our own snapshot, which is read off each model vendor's page and dated per row, and re-read three of those vendors' pages the same day. Twelve of the fourteen rows hold. Six were confirmed against the model vendor's own page on the day of writing, including both halves of DeepSeek's peak and off-peak split, which the table gets exactly right down to the cached-read column. One row we decline to judge: a context-tiered Qwen rate we carry as a single figure rather than two.

The row that fails is instructive precisely because it fails cleanly. For one MiMo model the table prints figures that are exactly four times what the model's maker publishes today, on the input column, the output column and the cached-read column alike. A single mistyped digit is a typo; the same multiplier across three columns is a systematic one. What we can say is that the reference rate is four times the maker's published rate. What we cannot say is that the quota drains four times faster, because the vendor never states that the meter uses these numbers — it says the table helps you understand the measurement, and the prose has to stop where the vendor stopped.

The other near-miss turned out not to be a miss at all, and it is the most useful thing on this page. The cheapest line on the rate card is a contributor-tier model, priced at a tenth of what its standard tier costs per million input tokens — and the maker's own documentation explains the gap in one sentence: heavily discounted pricing in exchange for permission to use your prompts and completions to train future models. The standard tier of the identical model carries the opposite promise, that prompts and completions are not used for training. So the rate card is right to the cent, our snapshot was simply missing a tier, and the twelve-and-a-half-fold difference between two rows of one model is not a discount. It is a price paid in something other than money.

§ 04 / LEAVING

Cancelling is self-serve. Unused money does not come back.

Cancellation itself is unremarkable and works the way it should: notify before the next term begins, through the same route you signed up by or by writing to support, and access continues to the end of the period you paid for. The terms then close every door behind it. Fees are payable in advance, payment obligations are non-cancelable once incurred, and fees paid are non-refundable. The refund clause names the specific hope it is removing — no refund for unused credits, for unused time, for pre-payments, or for any portion of the service — and adds that any refund at all is at the vendor's sole discretion.

Read that next to a clause five sections earlier, because the combination is the part worth acting on: the vendor reserves the right to suspend or permanently terminate access with or without notice, with or without cause, and for any or no reason. A prepaid balance is therefore money held at somebody else's discretion with no promise of return, which is an argument for topping up in small amounts rather than annually. Unlike the two vendors in this section whose refund and cancellation articles contradicted each other, both clauses here agree, and both agree against the buyer.

One gap sits underneath all of it. The terms make a separate document — Pricing and Payment Terms — the governing text for prices and the order of payment, and incorporate it by reference. We could not find it published. Six plausible paths return 404, the sitemap's hundred and thirty-one URLs contain nothing resembling it, and the terms page links only to the privacy notice; the footer was checked too, since a vendor covered earlier in this section published its terms nowhere else. The terms themselves carry a modification date of September 2025, which is a year before the subscription they would have to govern went on sale.

Frequently asked.

What people search next.

Q · 01 What is the cheapest way to run Cline? +

A model on your own machine, and it costs nothing but electricity and memory: no account, no key, no network. The vendor documents three local runtimes and sizes them by RAM rather than by price. If you want a hosted model instead, the floor is the rotating free-model promotion, which needs a Cline account, is capped by a quota published in no unit, does not work through the API, and carries the docs' own note that usage may be used to help improve model performance.

Q · 02 Does the Cline gateway add a markup? +

The vendor says no — the pricing page sells inference at cost and the enterprise page contrasts itself with tools that add markup. We cannot verify that directly, because no per-model rate card for the gateway is published on the open web. The indirect evidence is reasonably good: the rate card the monthly pass does publish, for other vendors' standard prices, matched those vendors' own published figures on twelve of fourteen rows when we checked, with one row overstated by exactly four times.

Q · 03 Is the monthly pass worth it against paying per token? +

It depends on a number the vendor does not publish. The pass promises two to five times the usage the same money buys at standard rates, which is a ratio rather than a quantity — there is no token, request or dollar ceiling stated anywhere, and the three metering windows are visible only inside the dashboard. So the pass is a good deal if you code most days on the open-weight models it includes, and an unknowable one otherwise. The calculator on this site will do the break-even arithmetic once you can state an allowance; the vendor has not stated one.

Q · 04 Which models does the pass include? +

Two published answers that overlap on nine of fourteen. The marketing page names one set, the documentation's rate card prices another, and each contains models the other omits. Three of the models on the marketing list are ones their makers have since retired or scheduled for retirement, according to those makers' own pages and our dated snapshot — including one whose name is still accepted by its vendor's API but now silently serves a different model. Check the model selector in the client rather than either list.

Q · 05 Can I be billed anything beyond the monthly fee? +

The pass's own page says so without quantifying it: a processing fee may apply on top. Separately, the gateway is a prepaid balance rather than a subscription, so the two can be used together and are billed separately. And the terms allow the vendor to amend the fees charged for credits or for existing products at any time, with the change taking effect in the billing cycle after notice.

Q · 06 What happens to my credits if I stop? +

They stay with the vendor. The refund clause names unused credits explicitly among the things a cancellation does not get back, and the same document permits termination of access for any or no reason. There is no expiry schedule published for credits either way, which means the sensible posture is to treat the balance as spending money rather than savings.

Q · 07 Is there a team plan? +

No middle tier exists. The pricing page has exactly two columns — the open-source client at no charge for individual developers, and an enterprise tier priced as Custom behind a contact form — and everything a team usually buys, from centralised billing to role-based access and audit logs, sits in the second one. The monthly pass is an individual product and is not on the pricing page at all.

Q · 08 Do I need a Cline account to use Cline? +

Only if you want the vendor's own inference. A local model needs nothing. Your own API key needs nothing from Cline either, and that route is the one its terms describe as leaving your tokens and your code entirely outside its systems. The account becomes necessary for the gateway, the monthly pass and the free-model promotions — which is also where the content licence in section 3.2 of the terms starts to apply.