TokenMix Research Lab · 2026-04-24

Trae IDE with Claude: Setup + vs Cursor 2026

Trae is ByteDance's AI-native IDE — effectively a Cursor competitor with a free tier (key differentiator) and strong multi-model support including Claude Opus 4.7, Qwen3-Coder-Plus, and Doubao Seed 2.0 Code. Initially positioned for Chinese market, Trae expanded internationally through 2025-2026 with ~300K active developers by April 2026. This guide covers Trae setup with Claude, how it compares to Cursor Composer 2, Qwen3-Coder + Claude Code combos, and whether the free tier is genuinely useful vs "freemium with limits." TokenMix.ai routes any model Trae supports through standardized OpenAI-compatible endpoint.

Confirmed vs Speculation
What Trae Actually Is
Setup with Claude Opus 4.7
vs Cursor Composer 2
Qwen3 Coder + Claude Code Combo
Free Tier vs Paid
FAQ

Confirmed vs Speculation

Claim	Status
Trae IDE developed by ByteDance	Confirmed
Has genuine free tier	Yes — limited but usable
Supports Claude, GPT, Qwen, Doubao	Confirmed
VS Code fork	Yes (similar to Cursor)
Free tier feature-limited vs paid	Yes
300K active developers	Reported by ByteDance
ByteDance's strategic answer to Cursor	Yes

Snapshot note (2026-04-24): The "300K active developers" figure is ByteDance-reported — not third-party verified. Free tier limits (500 req/day, a few Claude calls/day) reflect the current quota structure; ByteDance has revised Trae tier limits multiple times as they calibrate the free-to-paid conversion. Paid tier pricing ( 0-20) is approximate per region. Verify current terms on trae.ai before relying on a specific quota number.

What Trae Actually Is

Trae is:

VS Code-based IDE (fork with extensions)
Multi-model (Claude, GPT-5.4, Gemini 3.1 Pro, Qwen3-Coder, Doubao Seed 2.0 Code, DeepSeek V3.2)
Agent + chat modes — similar to Cursor
Free tier with quota (~500 requests/day)
Paid tier ( 0-20/mo) for higher quotas + exclusive models

Key difference from Cursor: no custom "Composer" model — Trae routes to upstream providers directly. No Trae-trained alternative.

Setup with Claude Opus 4.7

Install:

Download Trae from trae.ai
Install like any VS Code fork (preserves existing extensions from VS Code profile)
First launch: sign up or sign in

Configure Claude:

Settings → Models → Claude
Free tier: Claude access limited, use Anthropic key directly or via TokenMix.ai
Paid: Trae includes Claude in higher tiers

BYOK via TokenMix.ai:

Model Provider: OpenAI Compatible
Base URL: https://api.tokenmix.ai/v1
API Key: your_tokenmix_key
Model: anthropic/claude-opus-4-7

vs Cursor Composer 2

Dimension	Trae	Cursor Composer 2
IDE base	VS Code fork	VS Code fork
Default model	Configurable	Composer 2 (proprietary)
Free tier	Yes (500 req/day)	Limited
Monthly cost	0-20 paid	$20 paid
Multi-model support	Native	Via paid, limited
Claude Opus 4.7 support	Via BYOK	Via BYOK
Chinese market integration	Best-in-class	Good
International polish	Good	Better
Agent/workflow mode	Yes	Yes (more mature)
Offline features	No	No
Open source	No	No

Pick Trae if: free tier matters, Chinese market product, multi-model flexibility priority, ByteDance-ecosystem adjacent.

Pick Cursor if: Composer 2's IDE-optimized model quality matters, international polish, mature agent workflows (Cursor Composer 2 review).

Qwen3 Coder + Claude Code Combo

A specific Trae workflow: use Qwen3-Coder-Plus for inline completions (fast, cheap) + Claude Code for complex multi-file refactors (high quality).

Setup:

In Trae: default model = qwen/qwen3-coder-plus for inline
Parallel: run Claude Code in terminal for complex tasks
Or: Trae's agent mode triggered manually for multi-file work

Cost at typical usage (50 inline requests/hour + 5 agent tasks/day):

Trae Qwen3-Coder inline: ~$5/month
Claude Code Pro ($20/mo for complex): $20
Total: $25/month vs $20 Cursor Pro alone

Trade: 25% more cost for meaningful quality boost on complex tasks + Trae's free-tier buffer.

Free Tier vs Paid

Free tier genuine usefulness:

500 API requests/day (enough for solo dev daily work)
Access to Qwen3, Doubao, DeepSeek models
Basic agent workflows
Inline completions always on

Free tier limits:

Claude Opus 4.7: limited (a few requests/day)
GPT-5.4: limited
Advanced agent features: paid only
Offline / enterprise features: paid only

For solo devs on personal projects, free tier can be sufficient. For production coding, paid tier ( 0-20/month) unlocks full model access.

FAQ

Is Trae safe to use for US enterprise?

ByteDance association creates procurement scrutiny, similar to Doubao / Seedance. For US federal / defense: avoid. For most commercial use: fine. See OpenAI/Anthropic/Google vs DeepSeek for context — ByteDance not directly named but adjacent scrutiny.

Does Trae have its own coding model like Cursor Composer 2?

No. Trae routes to upstream providers (Qwen, Claude, GPT, etc.). This is different philosophy: ByteDance bets on multi-model routing rather than building custom IDE model.

Can I migrate from Cursor to Trae preserving settings?

Somewhat. VS Code extensions transfer. Cursor-specific features (Composer 2, BugBot, Ghost Mode) don't — Trae has its own similar-but-different versions. Budget a day to reconfigure.

Is Trae's free tier sustainable?

ByteDance subsidizes aggressively to capture market share vs Cursor. Sustainability depends on whether they can convert to paid. For 2026, free tier is reliable.

What about Qwen3 Coder as default?

Excellent choice. Qwen3-Coder-Plus at ~$0.40/ .60 is cheap and capable. Good balance with Trae's free-tier constraints.

Does Trae support Mac/Linux/Windows?

Yes all three major platforms. Native build for each; no WSL requirement.

Is there offline mode?

No — like Cursor and most AI IDEs, Trae needs internet for LLM calls. For offline coding, use local models via Ollama + Continue.dev VS Code extension.

Sources

By TokenMix Research Lab · Updated 2026-04-24