Complete archive of all blog posts from the Cline team
How Cline migrated its VS Code extension for 11 million developers to the new Cline SDK, built a safe A/B rollout, and reduced agent failures by 10x.
Benchmarking eight AI models on IMO 2026 problems in Cline, we found DeepSeek V4 Flash scored 30/42, clearing the gold medal cutoff for just $0.12. See how open-weight models compare with GPT-5.6 Sol, Claude Fable 5, Kimi K3, and more.
Build a code review agent with the Cline SDK using custom tools, guardrails, hooks, and a two-pass review workflow. Learn how to inspect PR diffs, filter false positives, audit tool usage, and post high-signal review comments to GitHub.
Learn how Cline evaluates and improves open weight coding agents using Terminal Bench, with practical heuristics for model performance, token efficiency, reasoning, providers, and eval optimization.
NVIDIA Nemotron 3.5 Lightning is now available in Cline, bringing a 30B MoE model with 3B active parameters, up to 1M context, and high throughput for agentic coding workflows.
Explore the best open weight models for coding in 2026. Compare Kimi K3, GLM 5.2, DeepSeek V4 Pro and Flash, and MiniMax M3 across benchmarks, pricing, model size, and developer use cases.
See how recursive self-improvement helped Cline optimize Kimi K3 and achieve an 88.8% SOTA score on Terminal-Bench 2.1 at a fraction of the cost.
A practical guide to the economics of self-hosting open-weight LLMs. Using Kimi K2.6 and real production traffic from Cline, this post breaks down GPU memory, inference, batching, pricing, and the point at which self-hosting can save millions.
ClinePass is a low-cost monthly subscription that pairs Cline's agent harness with a curated set of open-weight models and 2-5x the standard API rate limits, across every surface Cline runs on.
Learn how to extend the Cline agent loop with plugins and hooks to add custom behavior, lifecycle logging, and enforceable guardrails without rebuilding the harness.
0:00 /0:41 1× Before “agents” became a buzzword, Cline was the first real agentic coding experience. Cline started with the VSCode extension and helped a generation of developers step into AI coding. It was a great VS Code extension, but as the technology evolves, it also taught us something uncomfortable: the architecture that got us there was not the modular and extensible framework we would choose for what comes next. It was definitely the cost of b
An engineer at Cline told me he was vibe coding on a road trip from his phone. His Mac was at home running Cline agents; he was checking on them from the passenger seat. Here's the setup. What you need Your phone and your Mac both need Tailscale installed and signed in to the same tailnet. Tailscale is a free mesh VPN; once both devices are on it, your phone can reach your Mac's local services from anywhere. Step 1: Start Kanban on your Mac By default, Cline's Kanban board binds to 127.0.0
You're building on infrastructure you don't control, can't audit, and can't see degrading in real time. The Invisible Dependency For engineers, the inference vendor problem starts with a deceptively simple question: what happens when the model changes and you don't know about it? Most engineering teams treat their inference provider like a stable API. It isn't. Model versions rotate. Performance characteristics drift. Context window handling changes. Output formats shift in subtle ways that
We put together 20 starter prompts for the Kanban sidebar chat. They create linked dependency chains, maximize parallel agent execution, and produce real, working code. Install with npm i -g cline.
Here’s the elephant in the room about coding in 2026: the bottleneck isn’t the AI; it’s you. Not your skill. Not your prompts. Your attention. Your cognitive bandwidth. If you’ve spent any real time with coding agents, you know the feeling. You start the morning with a clean plan. Spin up a few agents. One is refactoring the auth module. Another is writing tests. A third is scaffolding a new API endpoint. You’re flying. Then around 10:30 AM, you look up and realize you have 20 terminal window
Why cheaper compute won't mean cheaper AI, and what your stack is risking right now. OpenAI projected operating losses of $74 billion in 2028 alone, before an expected pivot to profitability around 2030. Deutsche Bank research calculates that OpenAI's projected cumulative cash burn could exceed $200 billion by 2030 – making it the largest startup loss in business history. Every token you've sent at subsidized rates has been financed by venture capital and hyperscaler infrastructure deals, with
Every infrastructure engineer I know runs prompts in their head. They just don't call them that. You're deploying a new service to production. Without thinking about it, you follow a known sequence to ensure a service will run without impacting existing production workloads. You've done this dozens of times. You don't consult documentation. The checklist lives in your muscle memory, refined by every deployment that went sideways at 2am. Now think about what happens when you leave. Or when a ne
If you've built anything on top of AI models in the last year, you know the routine. Sign up for Anthropic's API, generate a key, store it somewhere safe. Do the same for OpenAI. Again for Google. Each provider has its own billing dashboard, its own SDK quirks, its own rate limits. Want to compare Claude against Gemini on the same task? That's two separate integrations, two sets of credentials, two billing statements. The Cline API eliminates all of that. One endpoint, one API key, access to mo
We didn't have benchmark numbers, so over a weekend we ran Cline against 89 coding tasks, diagnosed every failure, and shipped fixes that took our score from 47% to 57%. Here's the hill climbing process so you can do it too.
On February 17th, a compromised npm token was used to publish an unauthorized version of the Cline CLI. No malicious code was delivered, and the Cline VS Code and JetBrains extensions were never affected. Here's the full breakdown of what happened, what we found, and what we've changed.
Every few years, something shifts in how we write code and our workflows quietly adapt. Version control changed how we think about saving work. Code review changed how we think about readability. CI/CD changed how we think about testing. Now AI coding agents are changing something we barely think about at all: commit messages. Your commit history has a new audience. And that audience is reading more carefully than most of your teammates ever did. Here's what that looks like day to day. You ask
An open-source coding agent rebuilt for the terminal, with free models to get started.
MiniMax just released M2.5. It's available now in Cline across VS Code, JetBrains, Zed, Neovim, Emacs, and the Cline CLI. MiniMax is offering M2.5 for free for a limited time, so there's no barrier to trying it out.
Last week, hundreds of developers descended on Frontier Tower in San Francisco for ClawCon – and it was unlike any tech event I've been to. Laptop tattoos. Lobster merch flying off tables. The kind of energy that only happens when open source builders get together to celebrate something they actually care about. Cline co-hosted the event, and I was there representing us. Here's what happened. pic.twitter.com/5OC9tSZjhq — Dave Morin (@davemorin) February 5, 2026 What is ClawCon ClawCon is
Anthropic's most intelligent model is now in Cline. We've been testing Claude Opus 4.6 across the extension and the CLI. Here's why we're calling it The Deep Thinker.
I watched a developer waste $200 in API tokens one afternoon. Not because they were building something trivial – they were tackling important infrastructure work. Not because they were inexperienced – they had 15 years of backend development under their belt. The problem was simpler and more frustrating: they weren't wielding their tools effectively. Every few minutes, they'd hit revert. The AI would generate code that almost worked but missed the mark. They'd course-correct with another prompt
Redesigned terminal UI, better support for running parallel agents, ACP integration for Zed/Neovim/Emacs, and free Kimi K2.5 access (and more to come) during the release.
A premium car today contains over 100 million lines of code -- more than a fighter jet, more than Facebook's entire codebase. AI coding assistants promise to help close the productivity gap. But most get rejected before they reach a single developer's machine.
We surpassed five million installations of Cline across VS Code, JetBrains, Cursor, Windsurf, and other editors through OpenVSX.
Cline 3.55 adds two open models worth paying attention to. Arcee Trinity Large is free, US-built, and licensed under Apache 2.0. Kimi K2.5 is an open-source model that outperforms Opus 4.5 on certain benchmarks.
There's a growing assumption that AI coding assistants are fundamentally incompatible with regulated environments. This belief usually stems from how most AI tools are architected, not from AI-assisted development itself. Security constraints in these environments are real and non-negotiable. Restricted networks, strict compliance requirements, long approval timelines, and zero tolerance for opaque SaaS dependencies. Whether you're building software that handles financial transactions, processe
If you have an OpenAI subscription, you can now use it directly in Cline without managing API keys
Cline now offers comprehensive Jupyter Notebook support, giving data scientists the same AI-assisted experience that developers have come to expect – but designed specifically for how notebooks actually work.
Cline 3.51.0 is now available. This release adds support for OpenAI GPT‑5.2 Codex. OpenAI GPT‑5.2 Codex (gpt-5.2-codex) Cline 3.51.0 adds support for OpenAI GPT‑5.2 Codex (gpt-5.2-codex). OpenAI positions GPT‑5.2 Codex as an agentic coding model optimized for real-world software engineering: long-horizon work, big refactors/migrations, and reliable operation in terminal-heavy workflows. It also includes improvements in long-context understanding and tool-calling reliability, and stronger vis
Most AI coding tools require sending code through their servers. For healthcare systems protecting patient records, banks safeguarding transaction data, and defense contractors working with sensitive or classified code, that approach fails immediately. Security teams block the tool. Developers revert to manual workflows. Productivity drops, not because AI is ineffective, but because the architecture was wrong from the start. Cline takes a different approach. Your code never leaves your environ
Biweekly livestream every other Thursday at 9 am PT. One hour of Cline updates, industry expert guests, and Q&A.
Cline 3.48.0 adds Skills compatibility. If you've already built Skills, you can now use them in Cline. This release also adds websearch tooling through the Cline provider, giving Cline access to real-time information when you need it. Skills Think about how you'd onboard a new team member. You wouldn't dump every document on them at once. You'd give them an overview, then point them to detailed guides when they're actually working on specific tasks. Skills work the same way. Each skill is a
One of the most common pieces of feedback we hear: "I want to keep coding while Cline works." When Cline edits a file, it opens the diff view and takes your cursor along with it. That's fine when you're watching Cline work, but not ideal when you want to stay focused on your own code in another file. Introducing Background Edits Background Edits is an experimental feature that lets Cline edit files without opening the diff view or stealing your cursor. You stay in your file; Cline works in th
We're partnering with LG CNS to build “AI-Native Development” (AIND) an agentic AI solution that handles the entire development lifecycle from requirements to code.
What if you could get plain-English explanations for every line of AI-generated code? Meet Cline's Explain Changes feature: your AI slop detector, code review accelerator, and sanity-preservation tool rolled into one.
Faster response times, zero markup fees, and better global coverage. The Cline provider now runs on Vercel's infrastructure, delivering measurable performance improvements while staying true to our open source values.
GPT-5.2 brings frontier-level agentic performance. Devstral 2 is the best open-weights option for Cline. And the new model picker makes switching between them instant.
The stealth model is finally public: Devstral 2 scores 72.2% on SWE-bench Verified and works smoothly with Cline's multi-step tool workflows. Free for a limited time!
Understand AI-generated code before you ship it
A new agentic coding model built for Cline drops today with free access during alpha.
DeepSeek's new reasoning-first models are live. V3.2 for daily agentic coding; V3.2-Speciale for maximum reasoning power. Both at $0.28/$0.42 per million tokens.
Opus 4.5 is Anthropic's most capable model. Use it for complex multi-step problems, heavy MCP usage, and tasks requiring autonomous reasoning through ambiguity.
A call for contribution to establish reproducible, practical reinforcement learning environments sourced from real open source development work — with a $1M commitment to support open source maintainers.
Gemini 3 Pro extends what Cline can attempt as an autonomous coding agent while keeping you in control.
Cline 3.38.0 introduces Gemini 3 Pro Preview for enhanced reasoning and multimodal capabilities. We've also added AquaVoice Avalon, a voice-to-text model that specifically understands technical terms
After weeks with GPT-5.1, we discovered it's unlike any coding model we've seen. It investigates obsessively, follows structure religiously, and sustains focus across massive tasks. We adapted Cline to channel these traits into stable, long-running execution.
GPT-5.1 ships with Cline-optimized prompting. Plus Nous Research's Hermes 4 models, Avalon speech-to-text with Linux support, and critical bug fixes.
Hooks let you inject custom scripts at key workflow moments to validate operations, monitor usage, and shape AI decisions. Currently supported on macOS and Linux.
GitHub's Octoverse 2025 report just crowned Cline the fastest-growing AI open source project by contributors, with 4,704% year-over-year growth.
The rise of intelligent, tool-enabled coding agents marks a turning point in how software is written.
Native tool calling reduces errors and enables parallel execution. Auto-approve menu redesigned for better workflow. MiniMax M2 brings continuous reasoning, free until November 7.
Open source models can rival the best—if we give them the ground to stand on.
The model-agnostic, IDE-agnostic, inference-agnostic, open-source coding agent that 3 million developers trust—now with enterprise security, governance, and observability at scale.
Cline CLI Preview is available. After months of work, Cline is now a standalone service powered by Cline Core, exposing a gRPC API for scriptable automation, multi-instance orchestration, and building custom interfaces. The agent loop that works anywhere.
The Cline CLI (Preview) completes our vision of full agnosticism. One agent loop, Cline Core, that can run on any model, any surface, and any inference. The primitive for AI coding that enables everyone to build their own interfaces. Now available for Linux and macOS.
Analysis of Plan/Act model usage patterns from the last 7 days
Cline’s real-world telemetry shows Sonnet 4.5 at 96.2% and GLM-4.6 at 94.9% diff-edit success. The open-source gap is down to basis points. Try both in Cline.
After extensive testing, AMD found only specific models work reliably with Cline. Here's your complete guide to choosing the right model for your RAM.
Anthropic’s Claude Sonnet 4.5 is now in Cline, bringing 200k context, test-first workflows, and autonomous docs — built for real software development.
The stealth model gets 5x more context. Code-supernova now handles 1 million tokens.
If you're typing the same instructions to your agent every week, you're the process. Write them once as a markdown workflow, invoke with a slash command, and watch the agent execute with verification.
In v3.31, we're making Cline feel more natural to interact with. Three features that reduce the friction between what you want to build and getting it built.
OpenAI's GPT-5-Codex, optimized for coding agents, is now available in Cline with adaptive reasoning that uses 93% fewer tokens on simple tasks and 2x more on complex ones.
A mysterious coding model built for agentic coding, "code-supernova", drops in Cline with free access during alpha
How zAI's GLM Coding Plans change the economics of AI development -- $3/month gets you what otherwise costs $200
Cline is model agnostic, inference agnostic, and now platform agnostic -- running natively in JetBrains alongside VS Code. Your models, your providers, your IDE, no lock-in at any layer.
With LLMs multiplying fast, our new Fundamentals module helps you master model selection, tradeoffs, and coding workflows.
We’re launching AI Coding University to help developers build the core skills for agentic coding.
xAI extends free Grok access while we ship GPT-5 optimizations that actually matter for real coding tasks.
Most developers clutter their context with rules when they should build workflows. Workflows inject instructions on-demand and consume tokens only when invoked, while clinerules append to every system prompt.
Two mysterious models with 2M context windows appeared on major AI gateways. We tracked their performance across thousands of real coding tasks to see if the hype matches reality.
Kimi K2-0905 doubles the context window to 256k tokens and delivers some of the highest reliability tool calling we've seen in an open-source model. Built for coding agents that actually work.
For the first time, local models are powerful enough to run Cline effectively on a laptop.
hello! Three major capabilities land in Cline v3.26.6, each offering a different path to free AI-powered coding. Whether you want cloud speed, local privacy, or generous daily limits, this release has you covered. Speed without the wait xAI's Grok Code Fast model is now available in Cline, delivering responses so quickly that waiting for the model stops being a bottleneck in your workflow. Built specifically for agentic coding workflows, it excels at the iterative development sessions that d
Say hello to Grok Code Fast – or as you've known it, Sonic.
In building AI agents at Cline, we've discovered that the most dangerous ideas aren't the obviously bad ones, they're the seductive ones that sound brilliant in theory but fail in practice.
Free stealth model is now available in Cline
The Cline Context Playbook: when to use Deep Planning, Focus Chain, Auto Compact, /smol, /newtask, and Memory Bank
Bigger context windows don’t stop coding agents from drifting. Cline’s new Focus Chain keeps every step anchored to the goal — so even the longest tasks finish exactly where they’re meant to.
In v3.25, Cline never lets go of the thread.
Claude Sonnet 4 now supports 1 million tokens of context – a 5x increase from the previous 200K limit.
The most intelligent and reliable coding model from OpenAI is now available in Cline – and it's half the price of Sonnet 4.
Every developer knows the feeling. You're in flow, building something complex, and then – you wait. The model thinks. You watch the tokens slowly stream in. Your mind wanders. By the time the response arrives, you've lost your train of thought. Today, that changes. We're partnering with Cerebras to deliver code generation at 2,000 tokens per second in Cline. That's 40x faster than typical providers. No more waiting. Your thoughts flow directly into working code. How They Do It Cerebras didn
Cline has raised $32M in combined Series A and Seed funding, led by Emergence Capital and Pace Capital, with participation from 1984 Ventures
The age of AI coding subscriptions is ending. Not because companies want it to, but because the economics force it. When you can arbitrage a subscription in a commodity market, the market will take full advantage. Direct usage-based pricing is the only market-efficient outcome.
Test Kimi K2, the open-source model achieving 65.8% on SWE-bench. Available now in Cline.
Whether it be an Anthropic API key or using the Cline provider for inference, you can expect to pay what the provider charges for their inference, and none more. Cline doesn't charge any platform fees.
Andrei fixed the terminal! Cline v3.18.1-4 is a patch release series that addresses the terminal output issues that have been affecting your smooth usage of Cline.
Why hitting the reset button will actually save you time instead of fighting an upstream context current.
By connecting Cline to specialized servers for search, documentation, browser control, and more, you can create a workflow that gives Cline access to context outside of your codebase and your model's knowledge cutoff.
Cline v3.18 is a focused release that introduces the Gemini CLI as a provider, delivers significant performance and reliability upgrades for the Claude 4 family, and ships several important core improvements.
Tap into your existing Claude Max or Pro subscription to use Cline without paying per-message pricing.
When it comes to modifying code, Cline has two primary methods in its toolkit: 1. write_to_file for creating or overwriting entire files, and 2. replace_in_file for making surgical, targeted changes (diff edits). We call these targeted changes "diff edits," and their reliability is fundamental to the agent's performance. As part of our push to optimize every core subsystem of the agent, we've been intensely focused on improving the success rate of these diff edits. The result? We recently
We're excited to announce that Amazon Web Services has officially contributed 35 new Model Context Protocol (MCP) servers to the Cline ecosystem. This is a major milestone that extends the power of AI to nearly every corner of the AWS platform, allowing developers to manage their entire cloud infrastructure using natural language. For a long time, managing a complex cloud environment meant juggling dozens of consoles, dashboards, and configuration files. With these new MCP servers, that complex
For some time, Cline offered two ways to provide persistent guidance: a simple "Custom Instructions" text box and the more robust, file-based .clinerules system. While the text box was a familiar starting point, it was always a stepping stone. Today, we're fully embracing the more powerful paradigm by deprecating the old Custom Instructions feature to focus entirely on .clinerules. The reason is simple: treating instructions as code is fundamentally better. A single, static text box doesn’t sca
