Our take

The fastest-growing IDE of 2026 Q2 — 6M+ registered users, 98% Chinese-prompt accuracy, half the price of Cursor Pro for premium features.

What it is

Trae is an AI-native IDE shipped by ByteDance in 2025 and rolled out globally through 2026 Q2. Built on a VS Code fork, it supports one-click import of extensions/themes/shortcuts from Cursor or VS Code. Inside the IDE, Trae layers three agent modes — SOLO (end-to-end full app), Builder (project scaffolding), and inline prediction (CUE) — on top of a multi-model picker (Claude / GPT-4o / DeepSeek / Doubao-1.5-pro). It targets the same segment as Cursor and Windsurf but with a notably freer free tier.

Best for

Solo developers who want a free Cursor-like AI IDE
Coders migrating from Cursor / VS Code with zero setup cost
International teams needing multi-language IDE support (Chinese 98% accuracy)
Full-app builds from a single prompt

Pros

Free tier with 4 SOLO runs/day — no trial window limits
One-click import of Cursor / VS Code settings, themes, keymaps
Multi-model picker (Claude Sonnet 4, GPT-4o, DeepSeek, Doubao)
Strongest Chinese-language prompt accuracy on the market (98%)

Cons

Pro features (unlimited SOLO + premium models) cost $10/mo, similar shape to Cursor
Brand recognition outside Chinese markets still catching up to Cursor
Editor ecosystem (extensions, themes) narrower than Cursor/VS Code

Quick start

Go to trae.ai and download the IDE for macOS / Windows / Linux (English build).
Create an account with email or Google — free tier is unlimited on basic features.
On first launch, choose "Import from Cursor" or "Import from VS Code" to copy settings, extensions, themes, keymaps in one click.
Press Cmd+I / Ctrl+I to open inline AI chat; Cmd+L / Ctrl+L opens the Side Chat panel.
Switch model: in the AI panel, pick Claude Sonnet 4 / GPT-4o / DeepSeek / Doubao per task — switching is one click.
For end-to-end app builds, activate "SOLO Mode" from the top bar; describe the app, watch the agent scaffold it.

Sample input / output

Input

Project: small e-commerce site, Next.js 14 + Tailwind + Stripe test checkout

In SOLO Mode:
"Build a Next.js 14 storefront with: 6 product cards on the homepage, product detail pages, cart with Zustand, and a Stripe test-mode checkout. Use Tailwind. Add a /admin route stub. Deploy target is Replit."

In inline editor, hovering utils.ts:
"Add a TypeScript-native deep merge that preserves arrays (not Object.assign). Add tests."

Output

SOLO build (4m 18s on Trae Pro):
  Next.js project created at workspace/next-storefront
  Files touched: 38
  - src/app/page.tsx — 6 product cards, layout grid
  - src/app/product/[slug].tsx — detail page, add-to-cart
  - src/lib/cart.ts — Zustand store (8 actions)
  - src/lib/deepMerge.ts — TypeScript-native, preserves arrays
  - src/lib/deepMerge.test.ts — 4 test cases, all pass
  
  npm run build: passes
  4 npm warnings (deprecated packages) — noted, not blocking
  Deploy: "Live preview" turned on, Vercel-deployable

Cost: Free tier (4 SOLO runs/day) on basic; Pro $10/mo for unlimited + premium models.

Benchmarks

Funding / parent ByteDance, 2026 Q2 international rollout — Trae site

Models supported Claude Sonnet 4, GPT-4o, DeepSeek V3/R1, Doubao 1.5 Pro — Trae product

Plans Free (4 SOLO/day); Pro $10/mo annual; Teams custom — Trae pricing 2026

Users (cumulative) 6M+ registered — Trae site 2026

Chinese-prompt accuracy 9.8/10 (best in class) — CSDN 2026 review

One-click import Cursor / VS Code settings, extensions, shortcuts — Trae docs

Pricing

Free tier (4 SOLO runs/day); Pro $10/mo (annual) — unlimited SOLO + premium models + CUE Pro; Teams custom. International build runs Claude Sonnet 4 / GPT-4o / DeepSeek / Doubao-1.5-pro.

Underlying models

Claude Sonnet 4GPT-4oDeepSeek V3/R1Doubao-1.5-pro

Should you pick this?

Pick it if You want an AI-native IDE with a generous free tier, strong multi-language support, and the lowest friction from Cursor or VS Code (one-click import). Strong pick if you optimize for $/month over the top benchmark.

Skip it if You need the top SWE-bench scores (Claude Code is the leader) or you need the agent available on headless Linux / remote dev-server environments (Trae is desktop-centric).

Trae