Skip to content
Trae logo

Trae

ByteDance's AI IDE and standalone SOLO agent that scaffolds projects, runs terminal commands and tests in a browser, at $10 a month for Pro against Cursor's $20.

3.5/5 my assessment
Freemium
Software

Overview

ByteDance started Trae as a VS Code fork with an agent attached, then spent the following eighteen months making the agent the actual product. SOLO is the half worth caring about. You describe the outcome, it does requirements analysis, writes scaffolding, executes terminal commands, drives a browser to verify its own work and pushes a deploy. On 31 March 2026 it stopped needing the editor and shipped as a standalone desktop and web app.

Plans run Free, Lite at $3, Pro at $10, Pro+ at $30 and Ultra at $100 a month. February 2026 moved billing off request counts and onto tokens, which is more honest but much harder to budget: one long agentic loop over a big repository can eat a week of Pro allowance. Models come from multiple vendors with no key of your own required, and MCP support means your internal services can participate in the agent loop rather than sitting outside it.

For an engineer working from inside China this is the pragmatic default, because the alternatives depend on providers that are awkward to reach from the mainland. For anyone touching regulated data or client code, the defaults decide it: telemetry on, five years of retention, no full opt-out. Worth watching whether an enterprise tier with real data residency and deletion controls appears, because that is the only thing standing between Trae and adoption beyond hobby and prototype work.

Key Features

  • SOLO mode runs the whole loop: requirements analysis, scaffolding, terminal commands, browser testing and deployment
  • SOLO shipped as a standalone desktop and web app on 31 March 2026, no IDE plugin required
  • Frontier models from several vendors with no API key of your own
  • MCP support, so internal servers and tools sit inside the agent loop
  • Parallel cloud tasks, 2 concurrent on the free plan and up to 15 on Pro+
  • VS Code fork with settings and extension import, so switching costs an afternoon

Where it holds

  • Pro at $10 against Cursor at $20 with comparable model access. On capability alone the price gap is hard to defend.
  • The free tier is unusually generous for a product fronting paid frontier models, which makes evaluation cheap.
  • Pulling SOLO out into its own app is the smarter shape. Fewer places for context to leak than an agent wedged into an editor.

Where it breaks

  • Telemetry is on by default with five-year retention and cannot be fully disabled. Privacy Mode narrows the scope but does not remove it, which ends the conversation for client code under NDA.
  • Token-based billing since February 2026 makes monthly spend genuinely unpredictable once agents start looping over large contexts.
  • ByteDance ownership is a procurement problem in the US and increasingly in the EU, regardless of how the tool actually behaves.
  • Available models shift without much notice, awkward if you have tuned prompts and evals against a specific one.

My Take

The telemetry has to be settled before anything else. On by default, five-year retention and Privacy Mode narrows it without removing it, which rules the tool out for anything under NDA before a single feature gets discussed. Clear that hurdle and the value is hard to match: Pro is $10 where Cursor is $20, and SOLO has run multi-step work end to end since it went standalone on 31 March 2026. Good for side projects, internal tooling and prototypes. Much harder to defend for the repository your employer actually pays for.

Francis Okafor
Francis Okafor AI & Tech Lead · Engineer

Quick Info

Pricing:
freemium
Openness:
Proprietary
Starting at:
Free plan at $0 with rate limits and a capped SOLO allowance. Lite $3 a month, Pro $10, Pro+ $30, Ultra $100. Billing moved from fixed request counts to token consumption in February 2026, so long agentic runs burn allowance faster than the tier names imply.
Added:
Aug 2026
Updated:
Aug 2026

Use Cases

code generation software development agent development productivity

Judge it on your own work

The notes above say where Trae holds and where it breaks. The fastest check is your own workload.

Visit website ↗

Alternatives to Trae