Trae
ByteDance's VS Code-fork AI IDE with Builder and SOLO agent modes, a free tier plus token-based paid plans, and an open-source Trae-Agent CLI.
| Vendor | ByteDance |
|---|---|
| Form | AI-native IDE (VS Code fork) + open-source Trae-Agent CLI |
| Launched | January 2025 (China); founded 2024, HQ Beijing |
| ARscore | not scored |
| Pricing tier | Free tier + Lite $3 / Pro $10 / Pro+ $30 / Ultra $100 |
| Coding models | Doubao / DeepSeek / GLM / Kimi / Qwen / MiniMax (+ Claude on paid tiers) |
| Open source | CLI yes (MIT, trae-agent); IDE no |
Official links
Outbound, for trial and verification.
| Official site | https://www.trae.ai/ |
|---|---|
| Pricing | https://docs.trae.ai/ide/billing |
| English interface / changelog | https://docs.trae.ai/ide/new-plans-and-billing |
Capabilities
What it ships, in its own terms.
Builder and SOLO modes run multi-file generation end to end; MCP connects external tools; a Flow panel auto-switches editor, terminal, browser and docs. Switchable model roster including Doubao, DeepSeek, GLM, Kimi, Qwen and MiniMax, with Claude on paid tiers. SOLO became a standalone desktop and web app in March 2026.
Use cases
Scenario coverage (presence, not benchmark performance).
| Scenario | Local files | Docs / PPT | Excel reports | Analysis | Web search | Mini-programs | Web building |
|---|---|---|---|---|---|---|---|
| Trae | ✓ | — | — | ◐ | — | ✓ | ✓ |
✓ supported · ◐ partial or via a sibling product · ? unclear · — not in scope. Presence, not benchmark performance.
Users
Subscriber and active-user counts, source-labelled.
6M registered users and 1M+ monthly active in its first year — Source: Official (ByteDance annual report)
Third-party review
Independent coverage, not vendor marketing.
Third-party editorial: ★★★★☆ RECATOOLS 7.8/10 (capability 8, value 9, ease 8); Product Hunt 4.7/5
Coding ability (benchmark): not scored — no DeepSWE or Terminal-Bench measurement.
Independent teardowns (Unit 221B, The Register 2025) found heavy telemetry even when switched off; RECATOOLS rates it 7.8/10. No DeepSWE or Terminal-Bench score has been published.
Source: RECATOOLS / Press (The Register)
Compare further
The reference agents and the scored Chinese entry.