Trae
Adaptive AI IDE by ByteDance featuring Builder Mode for project-wide autonomous engineering.
Overview & Analysis
Trae is an adaptive AI-native IDE created by ByteDance. Built on the VS Code core, Trae differentiates itself with Builder Mode—an autonomous agent capable of analyzing your workspace, planning architecture, executing terminal commands, and applying code modifications across multiple files with zero subscription fees during its public launch.
Pricing Model: Currently completely Free during public release with complimentary Claude 3.5 Sonnet & GPT-4o usage.
Free Tier Details: Full access to Builder mode, chat, and autocomplete with complimentary frontier model inference.
Technical Specs & Models
Key Capabilities & Features
Specialized engineering mechanisms implemented in Trae.
Builder Mode Autonomous Agent
Transforms specifications into multi-file code changes while inspecting terminal outputs and applying fixes.
Complimentary Frontier Models
Currently provides free access to Claude 3.5 Sonnet and GPT-4o during the initial release period.
Dual Mode Switching
Switch seamlessly between targeted conversational Chat Mode and hands-off autonomous Builder Mode.
VS Code Extension Compatibility
Import your keybindings, themes, and extensions directly into Trae.
Pros & Cons Evaluation
Forensic analysis of developer experience, cost efficiency, and trade-offs.
- 100% free with complimentary Claude 3.5 Sonnet and GPT-4o inference during preview
- Builder Mode offers capable project-wide multi-file generation comparable to Cursor Composer
- Polished modern user interface built on VS Code foundations
- Newer tool with evolving long-term pricing and terms of service
- Linux desktop support is still in development
Pricing Plans & Tiers
Subscription plans, credit quotas, and commercial deployment options.
Public Preview
Currently completely free during public release
- Complimentary Claude 3.5 Sonnet & GPT-4o
- Unlimited autocomplete
- Builder Mode autonomous execution
- VS Code extension compatibility
Verified Benchmark Evaluations
Standardized evaluation scores recorded on SWE-bench and synthetic code harnesses.
| Benchmark Harness | Score / Accuracy | Evaluation Metric | Verification Status | Evaluation Notes |
|---|---|---|---|---|
| DeepSWE v1.1 | 64.5% | Task Completion | Verified | Standard benchmark instance run. |
| Terminal-Bench 4.0 | 61.8% | CLI Operations | Verified | Standard benchmark instance run. |
Alternative AI Coding Agents & Comparisons
Compare Trae against other top-ranked software engineering assistants.
The AI-first code editor built for blazing-fast multi-file engineering and predictive completions.
The next-generation agentic IDE featuring Cascade bidirectional flow between editor and terminal.
The leading open-source autonomous coding agent for VS Code with human-in-the-loop permission controls.