What is Trae?
Trae is an AI code editor from ByteDance, the owner of TikTok, run through its Singapore company SPRING (SG). It launched in January 2025 as a fork of Code OSS, the open-source core of VS Code, so your extensions and shortcuts mostly carry over. The editor (now called TraeCode) has an IDE mode for working alongside the AI and a SOLO mode for handing a task to an agent, and an Agent Window added in October 2026 runs several agents across projects, including cloud and scheduled tasks.
Paid plans give you a dollar allowance that is spent at each model's per-token list price. Which models you get depends on where you live: the US list includes Claude, GPT and Grok models, while the list for other regions has GPT, Gemini and Chinese models such as Seed, DeepSeek, Kimi and GLM but no Claude. You can add your own API key for any provider that speaks the OpenAI or Anthropic format. A business version, TRAE Enterprise, is sold through ByteDance's cloud arm, BytePlus.
How much does Trae cost?
| Plan | Price | What you get |
|---|---|---|
| Free | Free | Auto mode only, 5,000 autocompletions a month, 2 cloud tasks, no SOLO mode |
| Lite | $8/month ($80 annual) | $8 of model usage; sold in 12 countries only |
| Pro | $20/month ($200 annual) | $20 of model usage, choose any model, 10 cloud tasks |
| Pro+ | $60/month ($600 annual) | $60 of model usage, 15 cloud tasks |
| Ultra | $200/month ($2,000 annual) | $200 of model usage, 20 cloud tasks, early access to new models |
| Enterprise | Custom | Sold through BytePlus |
Free (Auto mode only, 5,000 autocompletions a month, no SOLO mode); Lite $8/month (12 countries only); Pro $20/month; Pro+ $60/month; Ultra $200/month. Annual: $80, $200, $600, $2,000. Each paid plan includes its price in model usage; On-Demand Usage after that is billed in $3 increments. Enterprise custom, through BytePlus.
Is Trae worth it?
If you write code for a living, your editor sees nearly everything you work on: your projects, your terminal, often your keys and passwords. So the question with any AI editor is not only whether it writes good code, but who is watching while it does.
Trae, from TikTok's owner ByteDance, does the everyday jobs you would expect. It finishes lines as you type, chats about your code, and lets you hand a whole task to an agent in SOLO mode. Since October 2026 an Agent Window runs several agents at once, across projects, including tasks that run in the cloud or on a schedule. Because it is built on the same open-source core as VS Code, moving over takes minutes.
Its old selling point was price, and that has gone. ByteDance told subscribers on 17 September 2026 that Pro would rise from $10 to $20 a month, Pro+ from $30 to $60 and Ultra from $100 to $200. The usage you get shrank too: according to Trae's own billing docs, an old Pro+ plan included $90 of model usage a month, and the new one includes $60. Existing monthly subscribers keep their old price as long as they never cancel.
The catch: the data story. In July 2025 a researcher found that Trae made around 500 calls to ByteDance servers in about seven minutes, up to 26 MB of data, with telemetry switched off. The Register reported that ByteDance responded by rewording the setting to say it only covers telemetry from the VS Code framework, not "other Trae tools", and did not say whether those can be turned off. At the time a Trae staff member said Trae does not train any AI models on user data. Trae's current docs say something different: your chats, including code snippets and AI replies, may be used for analytics, product improvement and model training unless you switch on Privacy Mode, which only works while you are logged in. Even with Privacy Mode on, Trae says some telemetry is still collected for basic functionality and performance. Trae's blog says data sits on servers in the United States, Singapore and Malaysia. Some of this is ordinary for the industry. The reversal on training, and the opt-out that does not fully opt you out, is not.
The model list is the other surprise. Outside the US you get no Claude models at all; you need your own Anthropic key for that. Paid plans are only sold in 59 countries and regions, a list that leaves out Canada and Ireland, and refunds are offered only within 24 hours and only if you have used nothing.
Our take: at $10 a month Trae was an easy recommendation for a student or a side project. At $20 it is a VS Code fork priced like its rivals but with fewer models than Cursor or GitHub Copilot, owned by a company that has already changed its story once about your data. On a personal project with Privacy Mode on, it is a reasonable editor. On your employer's code, read the privacy docs with your security team first.
Where it earns its place
A computer science student in India
Get an AI editor for coursework without paying US prices
The Lite plan, sold in India and 11 other countries, costs $8 a month or $80 a year and includes $8 of model usage plus SOLO mode. Turn on Privacy Mode under Settings > Account, and stay logged in, or your chats may be used for training.
A freelance web developer in the UK
Use Claude for agent work in Trae
The built-in list outside the US has no Claude models, so add an Anthropic key under Settings > Models using the Anthropic Messages format. Your Claude spend then goes to Anthropic; Trae's $20 Pro allowance covers only the built-in models, such as the GPT-6 family and Gemini 3.1 Pro.
A security lead at a 200-person software company
Decide whether developers may install Trae
Personal accounts fall under consumer terms with training on by default and telemetry that the in-app toggle does not fully stop. TRAE Enterprise is sold through BytePlus, whose product page lists SOC 2 Type 2; ask for the report and a written answer on which telemetry can be disabled before approving it.
Before you commit
- Setup
- Download the editor for macOS, Windows or Linux and sign in; it is a Code OSS fork, so VS Code extensions and settings carry over in a few minutes.
- Integrations
- Built-in models from OpenAI, Google, xAI and Anthropic (US) or OpenAI, Google, ByteDance Seed, DeepSeek, Kimi, GLM and MiniMax (elsewhere); custom models over OpenAI Chat Completions or Anthropic Messages APIs; Remote SSH; cloud tasks in the Agent Window.
- Data and privacy
- Chats, including code snippets and AI output, may be used for analytics, product improvement and model training unless Privacy Mode is on, and Privacy Mode only works while you are logged in. Codebase files are not used for training; they are uploaded to compute embeddings, then the plaintext is deleted and the embeddings kept. Some telemetry continues with Privacy Mode on. Data is stored in the United States, Singapore and Malaysia. Which AI tools train on your data?
- Cost at scale
- A 20-person team on Pro pays $400 a month, or $4,000 a year on annual billing, before any On-Demand Usage, which is billed per token at list price in $3 increments. Enterprise pricing through BytePlus is not published.
- Limits
- The Free plan is Auto mode only, with no model choice, 5,000 autocompletions a month and no SOLO mode. Paid plans are sold in 59 countries and regions only, and the model list differs between the US and everywhere else.
What’s good
- Built on VS Code's core, so extensions and settings carry over
- Agent Window runs several local, cloud and scheduled agents at once
- Bring your own key in OpenAI or Anthropic format for any provider
- Lite plan at $8 a month in 12 countries, including India and Brazil
- Existing monthly subscribers keep their old price
What’s not
- Prices doubled in September 2026, with less usage included on Pro+
- Chats may be used for model training unless Privacy Mode is on
- Some telemetry continues even with the setting off or Privacy Mode on
- No built-in Claude models outside the US
- Paid plans not sold in Canada or Ireland; refunds only within 24 hours
Best for
Individual developers on personal projects, especially in countries where the $8 Lite plan is sold, who are comfortable with ByteDance's data terms.
Trae alternatives
Trae: common questions
- Is Trae worth it?
- A capable VS Code-style editor that used to win on price; with prices doubled, what is left to weigh is ByteDance's handling of your data. We rate it 3.6 out of 5.
- How much does Trae cost?
- Free (Auto mode only, 5,000 autocompletions a month, no SOLO mode); Lite $8/month (12 countries only); Pro $20/month; Pro+ $60/month; Ultra $200/month. Annual: $80, $200, $600, $2,000. Each paid plan includes its price in model usage; On-Demand Usage after that is billed in $3 increments. Enterprise custom, through BytePlus.
- What are the downsides of Trae?
- Prices doubled in September 2026, with less usage included on Pro+; chats may be used for model training unless Privacy Mode is on; some telemetry continues even with the setting off or Privacy Mode on; no built-in Claude models outside the US; paid plans not sold in Canada or Ireland; refunds only within 24 hours.
- Who is Trae best for?
- Individual developers on personal projects, especially in countries where the $8 Lite plan is sold, who are comfortable with ByteDance's data terms.
- How much does Trae cost for a team or in production?
- A 20-person team on Pro pays $400 a month, or $4,000 a year on annual billing, before any On-Demand Usage, which is billed per token at list price in $3 increments. Enterprise pricing through BytePlus is not published.
- Is Trae safe for company data?
- Chats, including code snippets and AI output, may be used for analytics, product improvement and model training unless Privacy Mode is on, and Privacy Mode only works while you are logged in. Codebase files are not used for training; they are uploaded to compute embeddings, then the plaintext is deleted and the embeddings kept. Some telemetry continues with Privacy Mode on. Data is stored in the United States, Singapore and Malaysia.
- What are the best Trae alternatives?
- The closest alternatives we have reviewed are Cursor, GitHub Copilot, Kiro and Google Antigravity (with Gemini CLI and Jules).
Answers checked Nov 2, 2026.