Trae

Trae is an AI-first development tool with a desktop IDE, a web app and a mobile app. Its IDE window has a hands-on IDE mode and an AI-led SOLO mode. Plans are Free, Pro ($20 a month), Pro+ ($60) and Ultra ($200), billed by dollar usage. We checked the facts below on 31 October 2026 from Trae's own pricing page and docs. We did not test it ourselves.

At a glance

What Trae is

The docs open with a notice that TraeCode and TraeWork have been merged into "the all-new TRAE." That matters when you read older reviews and forum posts, because earlier material talks about separate products. Today the docs describe one product in three clients.

The "About TRAE" page calls it "a development tool deeply integrated with AI capabilities." You can stay in control of each step as you would in a normal IDE, or hand a task to an AI agent for planning and execution.

The three clients are:

Inside Desktop there are two windows. The Agent window is a task workspace for several agent tasks at once, including pull requests and hand-offs to the cloud. The IDE window is built around a code editor, supports VS Code extensions and split editing, and is where you run, debug and manage source control.

The IDE window also has two modes. IDE mode keeps the familiar editor, terminal, debugging and extensions. SOLO mode puts the AI in the lead.

It sits in the same family as Cursor, Windsurf and Kiro, which are editors with agents built in. SOLO mode also overlaps with app builders such as Lovable and Bolt, because it can go from a prompt to a deployed web app.

Who it suits

Trae suits developers who want an editor with strong agent features and who are comfortable choosing between models. The device-sync story helps people who start a task on a laptop and check it from a phone.

It also suits people who want something between a code editor and an app builder. SOLO mode takes a plain description, builds, previews and deploys, but you can drop back into the editor at any time.

It suits you less if you want a fixed, flat monthly price. Trae bills by dollars of usage, and heavier models burn through that allowance faster.

Pricing

Trae's pricing page lists four plans for most visitors. The docs page on plans and billing adds a fifth, Lite, with a short country list.

The pricing page marks Pro+ as "3x Pro" and Ultra as "10x Pro" on usage. It also shows a "Save 17%" switch for yearly billing, and the docs give the yearly figures above.

Other differences from the plans table in the docs:

• Queue priority: Free uses the standard queue. Every paid plan uses the fast queue.
• Autocompletion: limited on Free (5,000 a month on the pricing page), unlimited on paid plans.
• SOLO mode: the docs table marks it as not available on Free and available on Lite, Pro, Pro+ and Ultra.
• Early access to new models: the docs mark it for Ultra only. The pricing page shows "Limited" on the other plans and "Included" on Ultra.

The Lite plan is available only in Thailand, Indonesia, Vietnam, the Philippines, India, Pakistan, Turkey, Malaysia, Brazil, Colombia, Peru and Nigeria. If you are in the United States, the plan list is the four above.

How billing works

Trae charges by token use. Each model request in a chat or task is metered on its own, and the cost is token usage multiplied by the model's list price. Charges come out of your dollar allowance first. When the allowance runs out you can switch on On-Demand Usage, which keeps you working and bills actual consumption in $3 increments.

That means a plan's headline price and your real cost can differ. Cheap models stretch the allowance. Expensive ones shrink it. The docs publish a per-model rate table, covered below.

Refunds and payment

If you have not used any service within 24 hours of buying a subscription you can request a refund by email. After you have used the service, after expiry or after renewal, refunds are not available. Payment methods include major cards, PayPal (not in some countries), Alipay and WeChat Pay (not in the United States), plus local methods in a few countries.

Where paid plans are sold

The docs list the countries where paid services are sold. In North America it names the United States, Mexico and the Dominican Republic. Europe, Asia, Oceania, South America and Africa each have a longer list, including the United Kingdom, Germany, France, Japan, South Korea, India, Australia, New Zealand and Brazil. Check the list before you plan around a purchase.

What it builds and how it works

IDE mode

IDE mode is the editor experience. The docs describe a feature called CUE that gives code completion, chained completion, multi-line edits, next-edit prediction and navigation, smart imports and smart renaming in Python, TypeScript and Golang projects. You can attach context, use agents and use custom models.

SOLO mode

SOLO mode lets the AI plan and run the whole job: understanding requirements, writing code, testing, previewing and deploying. You can enter requirements as text, voice or uploaded files. The screen has a task panel, a chat panel and a tool panel.

The docs list these features:

• A built-in Agent that first writes a plan from your goal and project context, then builds step by step after you confirm.
• Task management for several tasks inside one project.
• Tool panels, including an editor, a documentation viewer and a browser.
• Figma to code, for whole frames or single components.
• A DiffView to review changed files and lines.
• Automatic folding of finished chat steps.

Integrations in SOLO mode

So the database, deploy and payment pieces are partner services, not a bundled backend. Compare that with Base44, which keeps those pieces inside one platform.

Custom agents and subagents

You can define custom agents with prompts, MCP servers and toolsets. Subagents are specialised agents defined as Markdown files with YAML frontmatter. The built-in Agent hands a task to a subagent when the description matches, and the subagent works in its own context window. User subagents apply to all projects on a device, and project subagents live in the repository under .trae/agents/. The feature sits under Settings, Beta, and must be switched on.

Models

The docs describe a built-in Auto model that picks a model for you by speed, performance and resource use. You can also choose a model by hand on paid plans.

The model page lists a long set with per-million-token rates, split into U.S. and non-U.S. lists. Examples from the U.S. list include:

The non-U.S. list on the same page includes models such as Seed-2.1-Turbo, GLM-5.2, DeepSeek-V4-Flash, DeepSeek-V4-Pro, Kimi-K3, Kimi-K2.7-Code and MiniMax-M3. It does not show Claude or Grok entries. The list is different by region, so check which one you see.

The docs say a group of newer models (GPT-6.1-Sol, GPT-6-Astra, GPT-6-Sol, GPT-6-Luna, GPT-5.6 variants, GPT-5.5, GLM-5.2, DeepSeek-V4-Flash, Kimi-K2.7-Code and Kimi-K3) is available only after you move to the new plans.

Custom models

You can add your own models with an API key. There are two routes: a preset provider list, or a custom setup where you give the API format (OpenAI Chat Completions or Anthropic Messages), a request URL, a model ID and a key. The TraeWork docs say custom models can only be added in the desktop client and run only in the local environment.

MCP support

Yes. The MCP overview says agents act as MCP clients and call tools on servers you add. Supported transports are stdio, SSE and Streamable HTTP. The documented protocol features cover requests, responses, timeouts, tool listing and calling, list-changed notifications, tool definitions with input and output schemas, text and structured results, and logging.

You add servers through Settings, MCP, either from a marketplace, by pasting JSON, or through install links. In the Agent window you pick a local or cloud runtime for each server. Local servers need NPX or UVX installed. Project servers live in .trae/mcp.json and need a switch turned on.

The docs add a disclaimer that MCP servers are third-party, and that Trae does not review or endorse them. Auto-run for MCP is off by default. With it on, the agent runs your servers and their tools without asking each time. All external MCP connections need your explicit approval the first time. For a primer, see what are MCP servers and the MCP tools directory.

Privacy and data handling

From the privacy pages:

• Chats, related code snippets and AI output may be used for analytics, product improvement and model training unless Privacy mode is on.
• Privacy mode only works while you are logged in. Logged out, it is off.
• Whatever the setting, Trae says it does not use your codebase files for analytics, product improvement or training.
• In the desktop client your codebase files stay on your device. For codebase indexing, files are uploaded temporarily to compute embeddings, then the plaintext is permanently deleted.
• With cloud services such as the web app and cloud agents, codebase files are processed and retained in the cloud so the service can run.

You switch Privacy mode on under Settings, Account. The pages we read do not list where data is stored by region, so check the privacy policy for that.

A TRAE account can be signed in on up to three devices at once. TRAE, TraeWork Desktop and TRAE Mobile count toward that limit, and the web app does not.

Limits and gotchas

• Product names changed. Old posts mention TraeCode and TraeWork. The docs say these merged into TRAE.
• Free has no SOLO mode according to the docs table.
• Usage is dollar-based. Premium models drain the allowance faster than Auto.
• Region matters. Paid plans are sold only in a listed set of countries. The model list differs by region.
• Custom models are local only.
• Auto-run is risky. The docs warn that command auto-run bypasses all security checks and runs outside the sandbox.
• Three-device cap on sign-ins.
• Free autocompletion is capped at 5,000 a month.

How it compares with peers

Our Cursor vs Windsurf and Cursor vs GitHub Copilot pages cover the established editors in depth. For pricing across the category see the coding agent pricing comparison. To pick between terminal and editor tools, read how to choose an AI coding agent.

Who should pick something else

• People outside the supported countries. Check the list first.
• Teams that need a flat, predictable price. Per-request pricing at model list rates is variable.
• Developers who want a terminal agent. Look at Claude Code or Gemini CLI.
• Non-coders who want hosting and a database in one place. Look at Lovable, Bolt or Base44.
• Anyone who needs written data-region commitments. The docs we read do not give them.

First-week plan

• Day one: install the desktop client and try IDE mode on a real repository.
• Day two: switch Privacy mode on and read what stays local.
• Day three: add one MCP server from the marketplace and keep auto-run off.
• Day four: try SOLO mode on a small web app if your plan includes it.
• Day five: compare the dollar cost of two models on the same task.
• Days six and seven: decide whether Pro's $20 allowance covers your week.

Our take

This is an editorial reading of Trae's pricing page and docs, not a hands-on test. The facts were checked on 31 October 2026 from the vendor's own pages.

The product has changed shape recently, and the docs show it: two separate lines became one, and plan names and quotas moved. We would trust the pricing page and the plans page over any third-party review older than a few months.

What stands out is the dollar-usage model. It is honest, since you see the per-model rate, but it asks you to think like a buyer of API tokens. The device sync and the cloud tasks are the real difference from a plain editor. If those fit how you work, the $20 Pro plan is the one to try first.

Checklist

• Confirm your country is on the paid-plan list
• Start on Free and see what Auto mode gives you
• Turn on Privacy mode while logged in
• Keep MCP auto-run off until you trust the server
• Check the model list for your region
• Watch usage in dollars, not requests
• Re-check the pricing page on the day you subscribe