How much Cline costs: the extension is free, you pay for tokens — a calculation
With your own key Cline sells no model subscription: you pay for tokens. A workday on several models, calculated, and how to stay within budget.
6 min · · экономия · IDE
If you are looking for the Cline price, there is no single number, because the extension and the models are different things. You can install Cline into an editor for free, but every step it takes sends a request to a model, and that request costs money. If your own key is connected, you pay the model supplier for tokens, with no subscription. Below we calculate what a workday and a month cost, how cache and model choice affect the bill and how to set a limit so you stay within budget.
What you pay for
Cline is an agent: it reads files, edits them, runs commands and decides what to do next from the result. Each such decision is a separate request, and in every request the agent sends the model the accumulated history of the task. So the bill is determined not by the number of your messages but by the number of agent steps and the length of the context.
About Cline's own terms: it has its own ways of accessing models (its own provider, subscriptions) with separate tariffs. We do not know what exactly they offer and for how much, and do not write about it; see the current information on Cline's site and in its documentation. "Cline for free" in the sense of "no fee for the extension" is true; in the sense of "models are free" it is not. Here we cover the own-key option: the calculation uses the gateway's prices.
Spend model: one workday
Assumptions for the calculation (adjust to your case):
- you give the agent 6 tasks a day;
- each takes 25 requests (agent steps);
- on average a request holds 40,000 input tokens (history and files) and 1,000 output tokens.
That is 150 requests a day: input 150 × 40,000 = 6,000,000 tokens (6M), output 150 × 1,000 = 150,000 tokens (0.15M).
Without cache
Zukko prices as of October 18, 2026, $ per 1M tokens. Formula: 6 × input price + 0.15 × output price.
| Model | Input / output | Formula | Day, $ | 22 workdays, $ |
|---|---|---|---|---|
deepseek-v4.1-flash | 0.013 / 0.038 | 6 × 0.013 + 0.15 × 0.038 | 0.084 | 1.84 |
glm-5.3-flash | 0.04 / 0.04 | 6 × 0.04 + 0.15 × 0.04 | 0.246 | 5.41 |
gpt-6.1-sol | 0.27 / 1.32 | 6 × 0.27 + 0.15 × 1.32 | 1.818 | 40.0 |
claude-sonnet-5-5 | 0.4 / 2 | 6 × 0.4 + 0.15 × 2 | 2.70 | 59.4 |
claude-opus-5-5 | 0.8 / 4 | 6 × 0.8 + 0.15 × 4 | 5.40 | 118.8 |
Checking one row: for gpt-6.1-sol, 6 × 0.27 = 1.62 and 0.15 × 1.32 = 0.198, total 1.818; times 22 gives 39.996, about $40.0.
For comparison, at the official catalogue prices claude-sonnet-5-5 (input $2, output $10) would cost the same day, without cache, 6 × 2 + 0.15 × 10 = $13.5.
With cache for Claude
The agent sends the same growing history again and again, so most of the input is repetition, and repetition can be read from cache at a reduced price. For Claude the discount comes from the Anthropic format (/v1/messages): the gateway sees how much input came from cache. In the OpenAI format (/v1/chat/completions) all Claude input is billed at the input price, that is, per the table above.
Assumption: of the 6M input tokens, 5.1M are read from cache, 0.3M are written to cache, 0.6M are fresh. Formula: fresh input × input price + read × read price + write × write price + output × output price.
claude-sonnet-5-5(input 0.4, read 0.05, write 0.5, output 2): 0.6 × 0.4 + 5.1 × 0.05 + 0.3 × 0.5 + 0.15 × 2 = 0.24 + 0.255 + 0.15 + 0.3 = $0.945 a day, $20.79 for 22 days.claude-opus-5-5(input 0.8, read 0.08, write 1, output 4): 0.6 × 0.8 + 5.1 × 0.08 + 0.3 × 1 + 0.15 × 4 = 0.48 + 0.408 + 0.3 + 0.6 = $1.788 a day, $39.34 for 22 days.
The difference from the table above is almost three times for Sonnet 5.5 (2.70 versus 0.945), but it follows from the chosen assumption about the cache share: yours may differ. The real token split is in the "Usage" section of the dashboard. A worked example for Claude Code is in the article on the cost of an agent with cache.
To send Claude through the Anthropic format, in Cline you can pick the Anthropic provider and set a custom address: then the Base URL is https://api.zukko.pro without /v1. With the OpenAI Compatible provider the address is https://api.zukko.pro/v1. Exact field names depend on the extension version, so check the Cline documentation and our VS Code extensions guide.
Two models: planning and acting
Cline separates Plan and Act modes and lets you assign different models to them (check this option in the settings of your version). Then the expensive model can be limited to planning and execution handed to a cheap one.
An example for the same day. Planning: 18 requests (3 per task) of 30,000 input and 2,000 output tokens on claude-opus-5-5 without cache: 0.54M × 0.8 + 0.036M × 4 = 0.432 + 0.144 = $0.576. Execution: the remaining 132 requests of 40,000 input and 1,000 output on deepseek-v4.1-flash: 5.28M × 0.013 + 0.132M × 0.038 = 0.06864 + 0.005016 = $0.0737. Total about $0.65 a day versus $5.40 working entirely on Opus 5.5 without cache.
This is arithmetic, not a promise of quality: whether a cheap model is good enough for execution, test on your own tasks. Strength indexes from the catalogue: claude-opus-5-5 58, gpt-6.1-sol 52, claude-sonnet-5-5 48, glm-5.3-flash 42, deepseek-v4.1-flash 39.
How to set a limit
The main protection against an unpleasant bill is not in Cline itself but on the key:
- In the API keys section create a separate key for Cline.
- Set a daily limit in dollars, say $3, and a monthly one, say $40.
- Restrict the list of allowed models (
allowed_models) so an expensive model picked by accident does not run.
When the daily limit is exhausted, the API answers with a 429 error and the code key_daily_limit_reached (monthly: key_monthly_limit_reached), and the extension shows it as a request error. That is a signal not to retry but to decide whether to raise the limit. How the limits work is described in the limits and cache docs.
How to keep spend down
- Start a new task for a new topic: long history is expensive at every step.
- Do not add the whole project to the context "just in case"; mention the files you need.
- On an unfamiliar codebase, check spend after the first hour, not at the end of the month.
- Check the auto-approve settings for a cap on consecutive requests: a looping agent is the fastest way to burn a budget (how to catch it).
- For GLM mind the 128K window: on long tasks it runs out sooner than on 1M models.
FAQ
Is Cline free or not?
Installing the extension costs nothing. You pay for the model: either through Cline's own access options, whose terms you should check with them, or through your own key, as in the calculation above.
Does Cline have a subscription?
With a Zukko key no model subscription is needed: a dollar balance is charged for tokens used. Which subscriptions Cline itself offers is current on its site.
How much does a workday in Cline cost?
In our example from $0.08 on deepseek-v4.1-flash to $5.40 on claude-opus-5-5 without cache; with Claude cache in the Anthropic format a day on Sonnet 5.5 comes to about $0.95. Your numbers depend on context length and the number of steps.
How do I pay from Russia?
Access goes through a gateway; the balance is topped up by SBP (the payment page adds an 8.5% fee on top), MasterCard or crypto. There is no subscription, you pay for tokens.
To check your own numbers, create a key with a daily limit in the Zukko dashboard and compare the first day's spend with the calculation above.
zukko
Pay 25×
less
for Claude Code & ChatGPT
One fetch_* key and base URL api.zukko.pro — for Cursor, Claude Code, Codex, and any SDK.
Related articles
- Claude Code with Qwen, GLM and DeepSeek models through one keyClaude Code is not tied to Claude: the model is set by a variable. A configuration for Qwen, GLM and DeepSeek, a session calculated, and how to keep Opus for the hard parts.
- opencode agents: setting up roles, models and toolsA custom agent in opencode is a role, a model and a set of permissions. A working config where a cheap model handles routine work and a stronger one edits code.
- Plan mode in Claude Code: plan first, then edit the codeIn plan mode Claude Code reads your code and proposes a plan but edits nothing until you agree. How to enable it, use it and what it costs.
- JetBrains AI Assistant not working: causes and workaroundsAI Assistant will not activate, stays silent or is off for some files. We go through the causes in the JetBrains docs and connect your own models via a gateway.