Trae Review, Pricing & Alternatives (2026)

Trae is an AI coding IDE made by ByteDance and launched internationally in January 2025. It combines an editor with chat, code completion, Builder, agent tasks, and SOLO workflows. ByteDance reported more than 6 million registered users across nearly 200 countries by the end of 2025, alongside plugin, IDE, and SOLO products and more than 100 feature updates. InfoQ described Trae as a fork of Microsoft's open source Code OSS editor, with Visual Studio Code plugin compatibility. ByteDance also reported nearly 60 million sessions and close to 500 million queries during 2025. Trae remains active, with current documentation listing Free, Lite, Pro, Pro+, and Ultra plans. The company publishes no funding figure or acquisition announcement.

Trae's workflow covers both project construction and routine editing. Chat can inspect project files, answer questions, suggest code, and help diagnose errors. In Builder or SOLO, a developer describes a feature, reviews proposed files and terminal actions, and checks the resulting preview. IDE mode handles manual edits and day to day completion. Developers can call built in or custom agents with @Agent, define coding or architecture rules, and add web or document context with #Web and #Doc. MCP integrations connect agents with supported tools. This makes Trae closer to an agent workspace than a completion plug-in, although the developer still needs to review edits, commands, dependencies, and generated output before merging changes.

Trae is built on the Code OSS family and combines editor workflows with chat, completion, Builder, and SOLO modes. Its April 2025 update unified Chat and Builder, introduced @Agent for built in and custom agents, added MCP integrations, and expanded context through Web, Doc, and Rules. ByteDance's December 2025 review says Trae shipped more than 100 feature updates, released the open source trae-agent project, and reached 10.2 thousand GitHub stars for that project. The company also reports support for up to 11,000 MCPs and 365,000 custom agents created or updated. The current billing page lists models including GPT-5.4, GPT-5.2, Gemini-3.1-Pro-Preview, and newer GPT-6 series entries, with availability tied to plan and upgrade status.

Trae's five public plans run from Free to Ultra. Free costs $0 and includes a limited monthly dollar allowance, auto mode, standard queue priority, limited autocomplete, and two concurrent cloud tasks. Lite costs $8 monthly or $80 yearly, while Pro costs $20 monthly or $200 yearly. Pro+ costs $60 monthly or $600 yearly, and Ultra costs $200 monthly or $2,000 yearly. Paid plans provide matching monthly dollar quotas, with 2, 10, 15, and 20 concurrent cloud tasks respectively. Each model request is metered separately. After an allowance is exhausted, On-Demand Usage can continue AI chat and bills actual token consumption in $3 increments. Paid services are unavailable in some countries, and Lite has a narrower regional list.

Trae's main tradeoffs are privacy, resource use, and metered billing. The company says codebase files are temporarily uploaded for embeddings and then deletes the plaintext uploads, while chat interactions may support analytics, product improvement, and model training unless logged-in users enable Privacy mode. TechRadar reported roughly 500 network calls and about 26MB transferred in seven minutes with telemetry disabled, plus higher memory use than Cursor. Cursor offers a more established contextual editing workflow, Windsurf centers agentic editor tasks, and GitHub Copilot fits existing IDE and repository workflows more directly. Trae is a sensible pick for developers who want natural-language application building, MCP support, and a low starting price, provided its data handling and usage billing fit the project.