Junie CLI pricing: plans, BYOK and API costs

Junie CLI itself has no price: installing it costs nothing, and JetBrains' own landing page advertises "Free to start. No card or subscription required." Every dollar you spend on it arrives through one of several separate routes underneath the client, and they do not bill the same way. The question "what does Junie CLI cost" therefore has no single answer until you say which route you are on — and two of the routes below are advertised as costing nothing at all.

One navigation note before any figure. Junie now has its own site and its own documentation tree: www.jetbrains.com/junie/ redirects to junie.jetbrains.com, and the old Junie licensing help page redirects into junie.jetbrains.com/docs/ (both observed as 301s on September 21, 2026). JetBrains AI Assistant is a separate product whose documentation did not move; it is cited here only for the AI Credit system the two share.

Junie CLI pricing: what actually bills you

Sources, all fetched September 21, 2026: Junie CLI, BYOK, custom LLM models, Junie Local and Junie CLI EAP. On the first row: the public JetBrains/junie repository is a distribution and issues hub rather than the CLI source, and its LICENSE.md points at the JetBrains AI Service Terms of Service — so this is a free-to-install proprietary binary, not an open-source one you could self-host. One route has no public price at all. The quickstart names a Junie API key with usage-based billing and points to junie.jetbrains.com/cli to generate one, but that URL is a sign-in wall with no rate card, no markup statement and no minimum. This guide quotes no figure for it because none is published. Note also that the CLI ships on two channels that disagree: the npm package @jetbrains/junie had latest at 3110.7.0, published September 9, 2026, while the release notes listed build 3196.4 from September 14 — so name the channel whenever you pin a version.

The two official price anchors disagree

JetBrains publishes the same plan names at two different monthly figures, and the help page carries two plan tables under one "Available license tiers" heading. Both readings are shown here rather than resolved into one number.

Read from the Junie landing page and the AI Assistant licensing and subscriptions page (last modified July 22, 2026), both on September 21, 2026. Do not budget from a bare figure here. The Junie page carries an Annual/Monthly toggle, and its credit allowances (10 and 35) match the help page's lower table exactly, which makes "the same plans quoted at two billing periods" the natural reading — but this guide could not capture the toggle state that produced $8.33 and $25.00, and found no "billed annually" small print, so that remains an inference. The help page's second table follows its discussion of organizations, which reads as a personal and an organization column; the page does not caption them in a way this guide can quote. Confirm the figure in checkout for your own account type before committing.

What the credit unit does settle is the unit of measure. The help page states that "Each AI Credit equals $1 USD charged in your local currency", that "The quota reset occurs every 30 days from the date when you first used your JetBrains AI license" — a rolling window, not a calendar month — and that top-up credits stay valid for 12 months. So an allowance is denominated in dollars of model usage, which makes it directly comparable with a prepaid balance. What is not published anywhere this guide reached is how many credits a given model or a given task consumes; the only mechanism statement is a September 2025 quota FAQ saying "AI usage isn't a flat fee because it depends on the type of model you use and the amount of text (tokens) flowing in and out" and "We're aligning usage to real, public provider prices per token". That post is about a year old, still compares "GPT-5" with "Claude 4", and describes the quotas it replaced as a temporary measure — so it is evidence the meter tracks tokens, and nothing more.

The surface boundary that decides whether you can buy your way out

This is the part a price table cannot show, and the part most likely to waste money. The Junie brand covers a terminal client, an IDE plugin, a GitHub Action and GitLab CI/CD, and they do not share the same answer about your own key. JetBrains states it flatly on its plugin models page: "Using your own API key with Junie is not supported via Junie plugin, however, you can do so in AI Assistant's AI chat in your JetBrains IDE." So buying a gateway key expecting it to light up Junie inside IntelliJ is a mis-purchase; BYOK and custom endpoints are Junie CLI capabilities.

The GitHub Action is a third case: it documents five named-provider secrets and says to supply either a Junie API key or one of them, not both. Custom JSON profiles are not mentioned on that page at all — treat that as undocumented rather than as a documented prohibition. And note what the custom-profile documentation does not say: the "No JetBrains AI subscription is required" sentence lives on the BYOK page and is scoped to the six named providers. The custom LLM page makes no statement about login or subscription either way, so this guide does not extend that sentence to cover custom endpoints.

JetBrains' model-selection page splits providers into four categories, and the distinction is the spine of any cost comparison: Junie models are "models accessed through a JetBrains AI subscription, by means of a Junie login or a JetBrains AI API key", BYOK models use "your own API key from a third-party provider such as OpenAI, Anthropic, Google, xAI, OpenRouter, or GitHub Copilot", Custom models are "models defined in a custom profile JSON file", and Proxy covers "models accessed through a custom proxy endpoint configured in config.json". A Kunavo key is not a BYOK key in this vocabulary — BYOK is that closed list of six named providers, and Kunavo is not on it; the route this guide documents is the Custom one, and JetBrains' Proxy category is a second way to reach an endpoint of your own that this guide did not test. Kunavo serves no embedding model, and text-to-speech and speech-to-text are not served here either, so a custom profile behind it carries Junie's chat traffic and nothing else.

Junie CLI API cost: a recomputable estimate

These figures are token arithmetic on stated assumptions — an estimate, not a measured task cost and not a bill ceiling. Assume one session sends 200k uncached input tokens and receives 12k output tokens. Rates are live Kunavo catalog prices per million tokens, and the last column asks how many such sessions $10 of prepaid credit covers — the same unit as an AI Credit allowance, which is what makes the two comparable at all.

Two readings worth having before you pick a plan. First, the spread between the cheap lane and the flagship lane is roughly five to one under these assumptions, so the model choice moves the bill far more than the route does — which is why a comparison that starts with plan prices starts in the wrong place. Second, Claude Sonnet 5 and Claude Opus 5 carry the same Kunavo rate on today's catalog, so at equal token usage the cheaper-model heuristic does not separate them at all — which of the two finishes your task in fewer attempts is a different question, and one only your own repository answers.

The helper traffic is the lever price tables miss. Junie's fasterModel role handles "internal helper tasks like summarizing context or classifying tasks", and role blocks "inherit the top-level properties" when not explicitly defined — so a one-model profile bills its housekeeping at your primary rate. Assuming that role adds 60k input and 4k output per session (an assumption for illustration; JetBrains publishes no share), it costs $0.160 on Claude Sonnet 5 against $0.032 when pointed at Claude Haiku 4.5. Two lines of JSON change that, and only id needs to differ because baseUrl, apiType and apiKey are inherited:

{
  "id": "claude-sonnet-5",
  "baseUrl": "https://api.kunavo.com/v1/chat/completions",
  "apiType": "OpenAICompletion",
  "apiKey": "${KUNAVO_API_KEY}",
  "fasterModel": { "id": "claude-haiku-4-5" }
}

One detail that costs an afternoon if you miss it: "The baseUrl is used as the complete endpoint URL — Junie does not append a path to it." That is the opposite convention from most clients, so the full path goes in, and changing apiType means changing the URL in the same edit. The four accepted apiType values are OpenAICompletion, OpenAIResponses, Google and Anthropic; Kunavo answers three of those formats and nothing at Kunavo matches the Google value. Also note that $${VAR} references in apiKey and extraHeaders are resolved against the environment when the profile loads, and that JetBrains documents the failure mode plainly: "If a referenced environment variable is not set, the profile fails to load and Junie reports an error identifying the missing variable." So an unexported key surfaces as a load error at startup, not as a failed request later.

Kunavo's catalog amount is a billing floor rather than a cap: when the upstream reports its charge, the bill is the greater of catalog cost and upstream cost times the applicable markup. Cache behaviour is excluded from this example on purpose, because nothing in Junie's documentation describes cache accounting for custom endpoints. The minimum top-up is $10 in prepaid credit — a funding minimum, not a task fee or a subscription. See billing details, and AI cost optimization for the measurement method.

Cheapest API for Junie CLI, and best API for Junie CLI

Take the two free floors first, because they beat every rate on the table above and most comparisons skip them. Junie Local is advertised as "Free to use. No card, no subscription, no credits" with "Nothing metered, nothing billed", running a fixed Qwen3.6-27B-4bit that is "fixed for now" with nothing to configure — behind a hard floor of an Apple Silicon M5 Mac, macOS above 26, 64 GB of RAM and about 40 GB of disk. The CLI Early Access Program is the other zero, free to join with a monthly quota at no cost, though the quota size is stated nowhere and joining runs through a JetBrains support form rather than a self-serve signup. Neither page states whether Junie Local is driven from Junie CLI or only from the IDE, so confirm that in your own build before planning around it.

Two boundaries on the "cheapest" question. Cheapest listed rate and lowest cost to finish the task are different claims: a model that needs three attempts at a low rate can cost more than one that lands first at a high rate, and only your own repository settles that. And effort level is a cost variable independent of model — JetBrains warns that "Higher effort levels do not always produce noticeably better results, but they can cost significantly more", and publishes no multiplier, so it cannot be folded into the arithmetic above. For the general comparison see best LLM gateway, OpenRouter alternatives and OpenAI-compatible API.

Set it up, then check the first charge

Kunavo publishes a setup reference for this client, and that is a published configuration, not a compatibility test: Junie CLI has not been installed or run against Kunavo's endpoint here, and every behavioural statement above is a documentation read. Keep a working route available while you try it, run one bounded task, then read the charge your account actually recorded — and treat Junie's own /usage readout with care, because whether it reports cost for custom-profile models or only for JetBrains-billed traffic is not established in the documentation. Start at the Junie CLI integration guide, and create a Kunavo account when you are ready to fund a key. Choosing between clients rather than routes? Junie CLI vs Claude Code compares the two agents directly.

FAQ

What was checked, on September 21, 2026: every source linked above was fetched that day — junie.jetbrains.com's landing page, its Junie CLI, BYOK, custom LLM models, model selection, plugin models, GitHub Action, Junie CLI EAP, Junie Local and What's New pages, the JetBrains/junie repository, the AI Assistant licensing and subscriptions page (last modified July 22, 2026), the September 2025 quota FAQ, and the npm registry entry for @jetbrains/junie. The two redirects were confirmed by request the same day. Nothing else was verified: no annual checkout was opened, no Junie API key rate was obtainable behind its sign-in wall, and the GitLab CI/CD variable names were not read directly and so are not printed here. Kunavo rates come from the live catalog at build time, and every dollar figure on this page is illustrative token arithmetic rather than a measured task cost. Junie CLI was not installed and no request was sent through it to Kunavo, so no claim is made here about streaming, tool calling or model routing inside the client.