Disclosure: some links on this page may be affiliate links; we may earn a commission at no extra cost to you. Read the disclosure.

By the toolist.ai editors. Researched from Cline's own pages and hands-on tests published by others, not hands-on tested by toolist.ai. Updated October 21, 2026.

Cline review: pricing, features and what testers found

Cline is a free AI coding agent, open source except its JetBrains plugin, that plans with you, edits files across your project and runs commands with your approval, in VS Code and its forks, JetBrains IDEs, the terminal, a desktop app (beta) and cloud sessions in the browser, on whatever model you connect. Cline itself costs nothing (you pay your model provider, or Cline credits that Cline says it sells at cost); its paid plan, ClinePass, costs $9.99 a month for quotas on a set of mostly open-weight models.

Our take

Cline suits developers who want an open-source coding agent they can point at any model: it works in VS Code and its forks, JetBrains IDEs, the terminal, a desktop app in beta and cloud sessions in the browser, lets you agree a plan in Plan mode before Act mode changes anything, and asks before each edit or command unless you turn on auto-approve. Cline itself is free: you pay your own model provider, buy Cline credits that Cline says it sells at cost, use a Claude or ChatGPT plan you already have, or run a local model. Its only subscription, ClinePass, costs $9.99 a month (Cline says a processing fee may apply and does not say how much) for quotas on the 12 models its docs list, mostly open-weight ones such as Kimi K3, GLM-5.3 and DeepSeek V4 Pro. The published tests date from spring 2026, and two of the three ran Cline on local models: The Register's Tobias Mann had Qwen3.6-27B one-shot an interactive solar-system web app and found the Plan and Act split a big help; XDA's Shekhar Vaidya got a seven-service homelab dashboard working with Devstral, but only after raising the context window for Cline's large system prompt and correcting a first run that reached none of the seven services; and XDA's Anurag Singh found Cline stable and safe in controlled VS Code workflows but less autonomous than Claude Code and slower in large projects. No cited test covers ClinePass, Cline's own providers, the desktop app, the cloud agents or the rebuilt agent harness Cline finished rolling out to its VS Code extension on August 23, 2026. The catches: what you pay your model provider is separate and depends on the model and how long the agent runs; ClinePass gives no quota figures beyond '2-5x' standard API rate limits over 5-hour, weekly and monthly windows; payments are non-refundable unless the law requires; and prompts sent through Cline's own providers can appear in its telemetry, which is on by default. This verdict rests on those three published tests, Cline's own pages and Meta's model docs.

Based on Cline's published plans and the hands-on tests linked below, not on a test of our own.

Good for

  • Developers who want an open-source agent with no Cline subscription and full choice of model, from a frontier API or a Claude or ChatGPT plan they already pay for to a model running on their own machine
  • Private or offline coding with local models through Ollama, LM Studio or Llama.cpp, as The Register's and XDA's Shekhar Vaidya's tests did, given a GPU or Mac with plenty of memory
  • Using open-weight coding models such as Kimi K3, GLM-5.3 and DeepSeek V4 Pro for a flat $9.99 a month without separate provider accounts or API keys (ClinePass)
  • Developers who want to agree a plan first and approve each edit and command inside VS Code or a JetBrains IDE, with checkpoints to roll back
  • Companies that want the agent on their own inference contracts (AWS Bedrock, Google Vertex AI, Azure OpenAI) with SSO and central controls, through Enterprise at a custom price

Watch out for

  • Cline is free, but the AI is not: with your own key every request is billed by your provider at its own token rates, so the cost depends on the model and how many turns the agent takes, and Cline publishes no estimate. Cline credits carry no markup by Cline's account ('at cost' on its pricing page), but its public pages list no per-model credit prices outside ClinePass's reference table. Running on a Claude Pro or Max plan instead draws on that plan's limits, and Cline's docs say image uploads and prompt caching are limited in that mode.
  • ClinePass publishes no quota numbers: usage is measured against a rolling 5-hour window, the calendar week and the calendar month, and the only figure is '2-5x the usage' of standard API rate limits. Cline's docs say its per-token reference prices show how usage counts against the quota, so a pricier model such as Kimi K3 ($3 input, $15 output per million tokens) uses it far faster than MiMo-V2.5 ($0.14, $0.28). Two of Trustpilot's three reviews complain about a paid plan's limits: one (September 2026) says the 5-hour limit allowed about 10 minutes of use, the other (October 2026) that one day used 60 percent of the monthly limit. The line-up changes: Cline's docs say GLM-5.2, Kimi K2.6, Kimi K2.7 Code and DeepSeek V4 Flash were withdrawn for capacity and upgrades, while the ClinePass page and model directory still list them.
  • ClinePass renews automatically every month until you cancel in your account settings before the renewal date; you keep access to the end of the paid month. Cline says an 'additional processing fee may apply' without saying how much, its terms say fees exclude taxes and are non-refundable (unused credits and unused subscription time included) unless the law requires otherwise, and free trials bill automatically the day after they end. No yearly price is published. The $4.99 first month that ran from launch on June 29, 2026 ended between August 31 and September 18, though the page still says $9.99 is the rate 'after promotion period'; Cline's launch post offers ClinePass for $1.99 through the Cline CLI 'for a limited period' and still says so, but no other Cline page we read mentions that offer, so whether it still applies is unclear.
  • Data depends on the provider: with your own key and your own infrastructure, Cline's terms say it does not receive your prompts or code. Through the Cline provider or ClinePass, your prompts and code pass through Cline to third-party model providers, which may use them for training unless you opt out where they offer it; Cline's telemetry for those requests can include the prompt and the model's response; and its terms license Cline to use your content to improve and develop its products, except while you hold a per-seat subscription. ClinePass's Muse Spark 1.3 Contributor has the name and reference prices of the Contributor mode of Meta's API, in which, Meta's docs say, your prompts, code and the responses may be used to train future Meta models. Cline's FAQ says it never trains models on code and prompts sent through its credits, while its free-models page says free-model usage may be used to improve the models.
  • Telemetry is on by default in the VS Code and JetBrains extensions, Cline Desktop and the CLI, even with your own keys: features and tools used, tasks, errors, token counts, model and provider names, your editor and operating system, a device identifier, and your account when signed in. You can turn it off (in the CLI by editing a settings file and stopping its background hub), after which Cline still contacts its servers for sign-in and feature flags.
  • Safety rests on your approvals: auto-approve's 'safe commands' are the ones the model itself marks safe, not a fixed allowlist, and YOLO mode approves everything, including file changes anywhere on your system and destructive commands. Cline is removing .clineignore because it never stopped access to files (shell commands could still read them). On February 17, 2026 an unauthorized party used a compromised token to publish Cline CLI 2.3.0 to npm, which for about eight hours installed the OpenClaw agent without users' knowledge; The Register, citing StepSecurity, put downloads at about 4,000, the fix is version 2.4.0 or later, and Cline now publishes through GitHub Actions with OIDC provenance.
  • Local models need serious hardware: Cline's docs suggest 16-32GB of memory for small quantized models and 64GB or more for larger ones, The Register recommends a GPU with 24GB of memory or a Mac with 32GB and warns that Cline's system prompt can bring weaker hardware to a crawl, and XDA's tester hit a context-window error until he raised Devstral's context from 8,192 to 32K tokens.
  • Cline's own pages disagree in places: the pricing page says 'no subscriptions', lists only Free and Enterprise (not ClinePass) and lists the JetBrains extension as an Enterprise feature, while the JetBrains plugin is free on the JetBrains Marketplace and the IDE page marks it early access. The Desktop app is a beta with frequent builds, Kanban is a research preview, and cloud agents work only with GitHub. The Teams plan the pricing page showed in December 2025 (free through 2025, then $20 a user a month with the first 10 seats free) is gone; teams now go through Enterprise at a custom price.
  • All three cited tests were published in April and May 2026, while Cline's extension was on its 3.x releases (two of them on local models, the third naming no model): before Cline 4.0 (June 26, 2026), ClinePass (June 29), the SDK-based agent harness Cline finished rolling out to its VS Code extension on August 23 (Cline says it cut tasks that hit its three-mistakes-in-a-row limit from 6.34% to 0.62%), Cline Desktop and the cloud agents. None reports using Cline's own providers, a frontier model through Cline, the CLI or the JetBrains plugin, and none names the Cline version tested.

What Cline does

Price

Free plan: Cline is free in VS Code and its forks, JetBrains IDEs, the terminal, a desktop app in beta and cloud sessions in the browser, and open source (Apache 2.0) except the JetBrains plugin, whose Marketplace page links no source code and names Cline's terms as its license; the editor extension needs no account. The AI is paid for separately: your own API key at your provider's rates, a Claude Pro or Max or ChatGPT plan you already have, Cline credits (which Cline's pricing page says it sells at cost), or a local model at no charge. A Cline account also gets rotating free models up to a limited quota, and Cline says usage of those free models may be used to improve the models.

PlanBilled monthlyBilled yearly, per monthIncludes
Open source (free) free not listed Cline in VS Code and its forks, JetBrains IDEs, the CLI, Desktop (beta), Kanban and the SDK, plus cloud agents in the browser; bring your own API key, a Claude Pro or Max or ChatGPT plan, or a local model, or pay with Cline credits; rotating free models up to a limited quota with a Cline account; community support
Cline provider (pay-as-you-go credits) not listed not listed Prepaid Cline credits on one balance across more than 100 models (300-plus by Cline's Desktop guide), with one sign-in and no API keys; Cline's pricing page says inference is sold at cost and the ClinePass page calls it pay-as-you-go API rates, with no per-model price list on Cline's public pages; no subscription; unused credits are non-refundable under Cline's terms
ClinePass $9.99 not listed Monthly subscription, renewing automatically: 12 models (Cline's docs list), mostly open-weight (Muse Spark 1.3 Contributor is Meta's hosted model), with quotas Cline puts at 2-5x the usage of standard API rate limits, measured over a rolling 5-hour window, the calendar week and the calendar month (no figures published); works in the IDE extensions, CLI, Desktop, cloud agents and SDK, and in other agents through a ClinePass API key; an additional processing fee may apply; no yearly price published
Enterprise not listed not listed Contact sales, custom pricing: SSO, an SLA and dedicated support, centralized billing, role-based access control, limits on which inference providers members use, a team management dashboard and authentication logs; runs on your own inference contracts (AWS Bedrock, Google Vertex AI, Azure OpenAI and others) at your negotiated rates with no Cline markup; OpenTelemetry export and prompt storage in your own S3 or R2 bucket

Prices last checked: 2026-10-30 on Cline's pricing page. Full breakdown: Cline pricing.

Price history

No price change since we first recorded these prices on October 21, 2026. We re-check them every week and list each change here.

What hands-on testers found

We have not tested Cline ourselves. These are the published tests we found that used it hands-on, newest first; follow the links for the full reviews.

XDA · Shekhar Vaidya · May 11, 2026

What they tested: Cline in VS Code with Devstral Small 2 24B (Devstral-Small-2-24B-Instruct-2512, GGUF) served by LM Studio on a PC with an RTX 4070 Ti and 32GB of RAM, with the Context7 MCP server, building a Flask dashboard for seven homelab services from the same detailed prompt he gave Claude Code. No Cline version is named; the test predates ClinePass, Cline 4.0 and the rebuilt harness and used no paid model through Cline. The page's only sponsored 'See at' links sit in a Claude product box (to claude.ai and chaton.ai); none goes to Cline.

  • Cline first stopped with a context-window error because Devstral was set to 8,192 tokens and Cline's system prompt is larger; raising the limit to 32K fixed it.
  • From the first prompt Cline made a six-step plan, wrote both files, installed the dependencies itself, ran tests and fixed a Jinja2 error it had introduced, yet the first run reached none of the seven services (wrong endpoints, missing authentication); one correction brought six of seven, and a couple more prompts all seven and the final design, where Claude Code needed five prompts in all.
  • He puts the gap down to Claude Code's built-in defaults and tooling rather than the model, found that without Context7 it took significantly more iterations, and calls the local stack a viable choice for privacy and cost if you accept more setup.

The Register · Tobias Mann and Thomas Claburn · May 2, 2026

What they tested: Cline's VS Code extension pointed at Qwen3.6-27B (Unsloth's Q4_K_M quantization) served locally by Llama.cpp on a 24GB Nvidia RTX 3090 Ti with a 65,536-token context, in a hands-on that set up Claude Code and Pi Coding Agent with the same model; Thomas Claburn's added section used Pi Coding Agent, not Cline. Small projects the authors themselves call fairly trivial; no Cline version is named, and the test predates ClinePass, Cline 4.0 and the rebuilt harness and used no paid provider through Cline. The article does not link to Cline.

  • A screenshot caption credits Cline running Qwen3.6-27B with one-shotting an interactive solar-system web app, and in their testing the model also found and patched bugs in an existing codebase (the article does not say in which agent); they add that a 27B local model is no replacement for frontier models.
  • They call Cline's separate planning and doing modes a huge help when you want to work through a problem rather than have the agent treat a question as an order.
  • They warn that Cline's and Claude Code's system prompts can slow weaker hardware to a crawl, and note that by default both wait for a person to OK edits and shell commands, unlike Pi Coding Agent.

XDA · Anurag Singh · April 5, 2026

What they tested: OpenCode, Cline and Aider over a few weeks as open-source replacements for Claude Code. He names no models, projects or Cline version, and the test predates ClinePass, Cline 4.0 and the rebuilt harness. The article does not link to Cline.

  • He found Cline built around VS Code, with diff views, approvals before commands and checkpoints that make it feel structured and safe, but tied to the editor: in larger projects or remote setups the extension can slow down, a friction he says becomes noticeable with constant use.
  • He rated it a reliable choice for VS Code users and more stable in controlled workflows, but less autonomous than Claude Code, so you guide it more often; OpenCode came closest to Claude Code in his test, and Claude Code stayed ahead of all three.

What users say

Test result

toolist.ai has not run its own test of Cline; the findings above come from the published tests linked. See how we research and test.

Best coding tools · Cline pricing

Do you make Cline? Claim this listing to correct it; it's free. To see whether ChatGPT, Claude, Gemini and Perplexity recommend it, and how many readers we send you: toolist.ai for AI companies.

Sources