Overview
Trae is an AI-native code editor launched by ByteDance, positioned as a domestic alternative to Cursor. It is built on VS Code, featuring AI code completion, conversational programming, and Agent (SOLO) mode, with deep integration of the Doubao large model and several mainstream large models.
**Note: Trae is no longer a completely free product.** Starting from 2026/02/24, Trae has fully switched its billing method from "per-use billing" to "**per-Token billing**", simultaneously launching five subscription tiers: **Free / Speed Pro / Speed Pro+ / Speed Ultra / Express**. The free version has no feature limitations and can continue to use mainstream models, but has the lowest cloud task concurrency (only 2); paid tiers mainly unlock higher "queue-skipping speed passes" and more cloud task concurrency, with Ultra/Express tiers also prioritizing access to SOTA models.
Key Features
- Five-tier subscription system: Free (¥0) / Speed Pro (¥59-99/month) / Pro+ (¥239/month) / Ultra (¥699/month) / Express (¥1399-1999/month), choose according to needs
- Free version features not reduced: All core features available in the free version, mainstream large models can still be used, only cloud task concurrency is the lowest (2)
- AI code completion: Real-time code completion and multi-line suggestions
- Conversational programming + SOLO agent: Generate code through conversation with AI in the editor; SOLO mode can autonomously analyze requirements, search code, and modify multiple files
- Chinese-friendly: Chinese conversation and interface, friendly to domestic developers
- VS Code compatibility: Built on VS Code, can import extensions and settings
Use Cases
- Developers with limited budget who only need the free version for daily programming
- Domestic developers looking for an AI programming tool
- Heavy users who need higher cloud task concurrency and are willing to pay for speed passes
- Students and beginners learning AI-assisted programming
- Teams that want to use AI programming in a Chinese environment
Pros
- Free version features not reduced: core AI programming capabilities available for free, low entry barrier
- Five-tier subscription covers light to heavy users, pay-as-you-go more flexible
- Available domestically: no additional network configuration required
- VS Code compatible: low migration cost
- Good Chinese experience: interface and conversation support Chinese
Pricing
**Free version**: ¥0, core features unlimited, cloud task concurrency 2. **Speed Pro**: New users limited-time ¥59/month (original ¥99/month), 100 queue-skipping speed passes per month, cloud task concurrency 10. **Speed Pro+**: ¥239/month, 300 queue-skipping speed passes per month, cloud task concurrency 10. **Speed Ultra**: ¥699/month, 1000 queue-skipping speed passes per month, cloud task concurrency 20, priority access to SOTA models. **Express** (limited): Limited-time ¥1399/month (original ¥1999/month), unlimited queue-skipping speed passes during validity, concurrency 20, priority access to SOTA models. Enterprise version requires separate consultation with Volcano Engine.
Summary
Trae initially used "completely free" as its core selling point, but since 2026/02/24 it has switched to a per-Token billing five-tier subscription system—the free version is still usable with no feature reduction, making it a good choice for light users; however, if you need to frequently run cloud tasks, pursue a queue-skipping experience, or prioritize access to SOTA models, you need to evaluate whether the ¥59-1399/month paid tiers are worthwhile. Compared to Cursor Pro's fixed $20/month, Trae's tiered pricing is more flexible, but it is no longer a "zero-cost" solution.
Version History
- 小红书与北大开源 UltraEP：面向大规模 MoE 训推的实时负载均衡方案 (2026-07-20): Xiaohongshu and Peking University have proposed UltraEP, which for the first time introduces real-time load balancing based on precise routing information into production systems, dynamically replicating hotspot experts at each microbatch and layer. On models such as Qwen3-235B, the training throughput achieves an average of 94.6% of ideal performance, a 42% improvement over Megatron-LM; inference prefill throughput is 1.56 times higher than SGLang.
- Per-Token billing + five-tier subscription (2026/02/24): Fully switched from per-use billing to per-Token billing, launched Free / Speed Pro / Pro+ / Ultra / Express five-tier subscriptions, ending the completely free era
- Trae 1.x / Trae Cloud (2026): ByteDance's AI IDE: Builder/SOLO mode automatically completes feature development, multiple models available, monthly active users exceeded one million
- Trae early (2025): ByteDance's first independent brand AI IDE, evolved from Marscode / Cody, completely free in early stages