Trae Review 2026: ByteDance's Free AI IDE, Honestly Tested

Some links on this page may be affiliate links. We may earn a commission at no extra cost to you.

Most Trae reviews were written at launch, when the headline was "unlimited Claude 3.7 Sonnet, completely free." That was true for about two months. In February 2026, Trae moved to a token-based freemium model — and most of those reviews still haven't caught up.

This review answers the two questions that actually matter for beginners right now: what does the free tier include in May 2026, and should you hand your code to a ByteDance product? Everything else follows from there.

If you want to see how Trae compares head-to-head with Cursor on pricing and agentic features, that full breakdown lives in the Trae vs Cursor comparison. This article focuses on the Trae experience itself. For context on where Trae sits among other tools that cost nothing, see the best free AI coding tools for 2026.

What Is Trae?

Trae is an AI IDE built by ByteDance — the same company that owns TikTok. It launched in early 2026 as a VS Code fork with deeply integrated AI features: a chat panel, inline completions, an agentic "Builder Mode," MCP support, and an in-editor web preview. The pitch was "everything Cursor does, but free."

Built by ByteDance — what that means for a coding tool

ByteDance is a well-resourced company with genuine AI investment, so the product quality is real. The relevance for most users is on the privacy side, which we cover in its own section. For the IDE itself, ByteDance's backing means Trae gets active development and isn't going away quietly.

VS Code fork vs extension: why it matters for setup

Trae is a standalone fork of VS Code, not an extension you install into an existing editor. That means you run it as its own application, separate from your regular VS Code. The upside is tighter AI integration. The downside is that your existing VS Code settings, keybindings, and installed extensions don't carry over automatically — you start fresh and re-import what you need. Extensions from the VS Code Marketplace and manually installed .vsix files both work.

What the Free Tier Actually Includes in 2026

Trae's free tier still exists, but it is not the "unlimited everything" tier from launch. Here is what it looks like as of May 2026.

Which AI models are on the free tier right now

The free tier gives you access to Claude Sonnet 4, DeepSeek R1, and GPT-4o — though under a request quota rather than the unlimited access from launch. At launch, Claude 3.7 Sonnet was on the free tier with no limits. Post-February 2026, the model roster upgraded to Claude Sonnet 4 and access shifted to a quota system: free users get 10 fast requests and 50 slow requests per month on premium models, plus 5,000 autocomplete completions per month.

Request quotas and what triggers the wall

Trae's free tier operates on a monthly request quota: 10 fast requests and 50 slow requests on premium models per month, plus 5,000 autocompletions. Once you hit the monthly limit, the editor continues to work — you just lose AI responses until the quota resets or you upgrade. The wall is a soft one: you can still write code, you just lose the AI layer.

Builder Mode on free: what you get and what's cut

Builder Mode — Trae's agentic scaffold feature — is available on the free tier with usage governed by the same request quota. Heavy Builder Mode sessions consume requests faster than simple chat queries, so a complex project scaffold can eat a meaningful chunk of your monthly allowance.

Trae's Paid Tiers: Lite, Pro, Pro+, and Ultra

Trae introduced paid tiers in February 2026. The current lineup is Lite, Pro, a Pro+ tier at $30/month, and Ultra at $100/month. Each tier increases the request quota and unlocks higher-tier model access.

What each tier adds over free

• Lite: Higher request quota than free; full access to premium models including Claude Sonnet 4
• Pro: Larger monthly usage allocation plus unlimited standard queue requests; eliminates the queuing that free users experience under heavy load
• Pro+ ($30/mo): Bigger usage quota than Pro, aimed at developers running frequent AI code generation and agent calls
• Ultra ($100/mo): Highest quota tier; priority access to the latest models and the most concurrent cloud tasks in Trae Solo; aimed at power users and small teams running heavy agentic workloads

Is the entry-level Lite plan worth it for beginners?

For a beginner building personal projects, Lite is the tier to consider if you hit the free wall. It's the cheapest paid plan and meaningfully expands what you can do with Builder Mode without jumping to a heavier commitment. Lite unlocks full access to Claude Sonnet 4 and removes the tight monthly cap that makes the free tier feel limited once you start real projects.

How it stacks up against Windsurf Free and Cursor Free

At the free tier, Trae competes most directly with Windsurf. The Windsurf review walks through what Windsurf's free tier includes — the core tradeoff is that Windsurf's free tier offers autocomplete without the ByteDance question, while Trae's free tier (when models are available) gives you more capable agentic features. Cursor's free tier is more limited still. For a full three-way pricing breakdown, the Trae vs Cursor comparison covers it directly.

Core Features Walkthrough

Builder Mode: natural language to full project scaffold

Builder Mode is Trae's most distinctive feature. You describe what you want to build — "a to-do app with a dark theme and local storage" — and Trae generates the full file structure, writes the code, and drops it into your workspace. It handles multi-file edits and iterative changes through conversation, similar to Cursor's Composer or Windsurf's Cascade.

For beginners, this is the feature that makes Trae genuinely useful rather than just a slightly smarter editor. You can go from zero to a running app without knowing exactly what files to create or how to structure the project. If you want to see how Builder Mode output compares to dedicated app builders, the best AI app builders comparison covers that territory.

Inline completions and Chat Mode

Inline completions work the way you'd expect from any modern AI IDE — ghost text appears as you type, you hit Tab to accept. Chat Mode is a persistent panel where you ask questions, request edits, and have the AI explain code. Neither feature is novel at this point, but both are competent.

MCP support

Trae supports the Model Context Protocol (MCP), which lets you connect external tools — file systems, APIs, databases — directly into the AI context. MCP is available on all tiers including free; Trae does not gate MCP connectivity behind a paid plan. You connect any MCP-compatible server through the settings panel, the same workflow as Claude Code or Cursor.

In-editor web preview and Figma-to-code

Trae includes a split-pane web preview so you can see a running version of your app alongside your code without switching to a browser. The Figma-to-code feature lets you import a Figma design and have Trae generate component code from it — a workflow that's genuinely useful for vibe coders who start with a design mockup rather than a blank file.

Extension compatibility: VS Code marketplace and .vsix files

Because Trae is a VS Code fork, the extension ecosystem is largely intact. You can install most extensions directly from the VS Code Marketplace inside Trae, and .vsix files install manually via the Extensions panel. A small number of extensions that rely on VS Code internals may not work, but the common ones — ESLint, Prettier, Tailwind CSS IntelliSense, GitLens — do.

The ByteDance Privacy Question

This section deserves a plain answer, not a buried disclaimer.

What telemetry Trae collects

Trae collects usage telemetry: which features you use, error logs, session data. More relevantly for developers, the AI features transmit your code context — the files and snippets you share with the AI — to Trae's servers for model inference. That is standard for any cloud-based AI IDE, including Cursor and Windsurf. The ByteDance-specific concern is about where those servers are and what legal jurisdiction governs the data.

Trae's privacy policy states that data may be retained for up to five years after account closure. That is a longer retention window than most comparable tools.

Whether you can opt out (and what happens if you can't)

Trae added a Privacy Mode toggle in settings that is described as disabling analytics and product-improvement telemetry. However, independent security researchers have found that telemetry transmission continues even when this toggle is enabled — Trae was observed making hundreds of network calls and transferring substantial data per session with the opt-out active. ByteDance acknowledges that some basic telemetry is retained for "functionality and performance monitoring" even in Privacy Mode. The toggle is better than nothing, but it does not fully stop data transmission.

Who this should and shouldn't actually concern

If you are building a personal portfolio site, a small side project, or learning to code for the first time, the privacy question is largely academic. Your code has no commercial value and no sensitive data. For that use case, Trae is fine.

If you are building anything with proprietary business logic, customer data, authentication flows for a real product, or code under an NDA — use a tool where the data governance is clearer. This is not a Trae-specific warning; it applies to any cloud-inferencing IDE. ByteDance's jurisdiction adds a layer of uncertainty that tips the balance toward caution for sensitive work specifically.

Getting Started with Trae

Download and install on Windows and Mac

Trae is available for Windows and Mac. Linux is not officially supported as of May 2026. Download the installer from trae.ai, run it, and Trae opens as a standalone application. Installation is straightforward — no terminal commands, no configuration files.

First run: choosing your model and connecting your account

On first launch, Trae prompts you to sign in or create an account. An account is required to use the AI features — you cannot use Builder Mode or Chat anonymously. After sign-in, you choose your default model from the available options. You can switch models per-session from a dropdown in the chat panel.

Switching between Claude Sonnet 4, DeepSeek R1, and GPT-4o

Model switching is done inline — there is a model selector in the Chat and Builder Mode panels. You pick the model before starting a conversation or a Builder Mode session. Switching mid-session starts a new context window. The available models depend on your tier; the free tier shows only the models your quota covers. Gemini 2.5 Pro is also available in the model roster depending on your plan.

Who Trae Is Actually For

Best fit: beginners, solo vibe coders, budget-first learners building non-sensitive projects

Trae is the right starting point if you are new to AI-assisted coding, you want to build personal or portfolio projects, and you want to spend as little as possible while having genuine agentic features available. The free tier — even in its post-launch limited form — gives you enough to learn the workflow. Builder Mode in particular makes Trae one of the better tools for vibe coding: describing what you want and letting the IDE construct the project structure for you.

The entry-level Lite tier makes Trae competitive with free alternatives when you need more headroom. At that price point there is almost no argument for paying more before you have validated that AI-assisted coding is part of your regular workflow.

Not a great fit: teams, proprietary codebases, privacy-strict environments

Teams need version-controlled AI context and shared settings — Trae is a solo tool right now. Anything involving proprietary code, client work, or data with compliance requirements should go through a tool with clearer enterprise data governance. And if you are on Linux, Trae is not yet an option.

Verdict

Trae is a real, capable AI IDE with a free tier that still has meaningful value in 2026 — it just is not the "unlimited everything" product from launch. Builder Mode works. The VS Code extension compatibility is solid. The model selection is competitive when your quota is healthy.

The ByteDance question is real and you should answer it for yourself before putting sensitive code through Trae. For personal projects and learning, it is not a blocker. For anything professional or proprietary, it deserves more scrutiny than a checkbox.

If the ByteDance angle makes you uneasy and free is the priority, the Windsurf review covers the clearest alternative. If you want to see exactly how Trae's pricing and agentic features line up against Cursor before deciding, start with the Trae vs Cursor 2026 comparison.

For a beginner who wants a capable AI IDE, does not want a monthly bill on day one, and is building projects where privacy is not a concern — Trae is genuinely one of the best starting points available right now.