Trae

Short answer. Trae is ByteDance’s AI-native IDE, split into two surfaces — TraeCode, a “10x AI coding engineer” for developers, and TraeWork, a “professional AI work assistant” for non-developers, with a browser-based web edition — competing in both the coding-agent and AI-workspace races under a single brand.

Key facts.

• Two surfaces under “Ship Faster with Trae”: TraeCode (coding agent, with IDE and SOLO autonomous modes) and TraeWork (work assistant, with a web edition at work.trae.ai).
• Framed as spanning both coding and general knowledge work, in contrast to coding-only agents in the database.
• The public marketing page is deliberately sparse: it states the product split and download paths but does not name the underlying model, spell out feature tiers or publish pricing.
• Closed, cloud-first product; no self-hosted or open-source edition documented.
• Sits in ByteDance’s agent portfolio alongside Coze and Doubao, but with a developer-IDE posture.

What this means. Trae is ByteDance’s entry into the coding-agent and AI-workspace race — a surface-expansion bet that extends the company’s agent portfolio from the consumer assistant and no-code platform into the developer IDE.

What is uncertain. The underlying model powering the coding agent, pricing tiers, feature depth, and whether TraeCode routes to ByteDance’s own models or third-party models are not documented on the public page.

Sources.

evidence_id	source_name	source_url	source_type	published	verified	confidence	conflict
src-agents-trae-1	Trae official site	https://www.trae.ai/	Official	—	2026-09-30	high	—
src-agents-trae-2	TraeWork web edition	https://work.trae.ai/	Official	—	2026-09-30	high	—
src-agents-trae-3	TraeCode IDE page	https://www.trae.ai/ide/	Official	—	2026-09-30	high	—

Why it matters

Trae matters as ByteDance’s coding-agent and AI-workspace bet. China AI Hub analysis indicates its structural role is surface expansion: ByteDance already has the consumer assistant (Doubao) and the no-code platform (Coze), and Trae extends that portfolio into the developer IDE and the general work assistant — the two surfaces where Alibaba (Qwen Code) and Moonshot (Kimi Code) already compete.

China AI Hub analysis: the thin public documentation is itself the signal. Unlike ByteDance’s model and API pages, which are richly documented, Trae’s public page reveals little about model routing or pricing — consistent with a product still consolidating its two surfaces rather than a mature, fully documented platform. The two-surface split (TraeCode for developers, TraeWork for knowledge workers) also means Trae competes on two fronts at once, diluting a clear single-surface identity in exchange for broader reach.

China AI Hub analysis: Trae is positioned as a product bet more than a platform bet — which is why its model routing and pricing sit behind the product rather than in public docs. For a buyer, that means evaluating Trae requires trying the product, not reading a spec sheet — a lower-transparency posture than ByteDance’s Ark API or Coze documentation.

How it differs from Kimi Code and Qwen Code

Trae is a coding and work product, against two coding specialists.

• Kimi Code is a coding agent bound to Moonshot’s K3/K2.7-Code models, with a documented 1M-token output ceiling from K3.
• Qwen Code is an open, multi-protocol coding agent (OpenAI/Anthropic/Gemini/Qwen plus DeepSeek/MiniMax/Z.AI/Kimi/local), notable for model choice.
• Trae is a closed, cloud-first IDE that splits into TraeCode (coding) and TraeWork (general work), with the underlying model undisclosed.

China AI Hub analysis: the contrast is transparency and scope. Kimi Code and Qwen Code document their models (bound to Moonshot, or multi-protocol), so a buyer can reason about the model ceiling; Trae documents neither model nor pricing publicly, so the buyer cannot — but Trae spans general knowledge work through TraeWork, which the two coding specialists do not. Trae competes on breadth of surface; the coding specialists compete on model transparency and depth.

Practical implications

For developers. TraeCode offers IDE and autonomous SOLO modes, but the model powering it is undisclosed — a developer cannot confirm the model ceiling from public docs and must try the product to evaluate it.

For non-developers. TraeWork and its web edition extend the same brand into general knowledge work — a surface the coding specialists do not cover — but with the same sparse public documentation.

For evaluators. Trae is a closed, cloud-first product with no self-hosted edition, so it is not an option for teams that need data control or model transparency; those teams must look to Qwen Code (open, multi-protocol) or Kimi Code (open, but Moonshot-bound).

What the evidence shows

The evidence for Trae is thin by design. The official site establishes the two-surface structure (“Ship Faster with Trae” leading to TraeCode and TraeWork downloads, plus a TraeWork Web edition) and the TraeCode IDE page documents the IDE/SOLO mode distinction — but the page stops there. No underlying model, no pricing tiers, no feature depth.

China AI Hub analysis indicates this is a transparency gap relative to ByteDance’s own other surfaces: Coze documents its Doubao default and its model node, and the Ark API documents its pricing and model tiers, but Trae documents neither. The asymmetry is consistent with a product still consolidating two surfaces, and it means the database can record Trae’s structure and positioning but not its model or economics — which is exactly the information a buyer most needs to reason about a coding agent.

Where this fits

Workload	Relevance
AI-assisted coding (TraeCode, IDE + SOLO)	High
General knowledge work (TraeWork + web)	High
Browser-based AI work (TraeWork Web)	High
Model-transparent coding	Low (model undisclosed)
Self-hosted / open-source deployment	None (closed, cloud-first)
Multi-protocol model routing	Not documented

Relevance judgments are China AI Hub analysis based on documented capabilities, not vendor claims.

Field reference

Field	Value	Evidence type
Underlying model	Not publicly documented	Not publicly documented
Target users	Developers (TraeCode); knowledge workers (TraeWork)	Official
Platform	IDE (TraeCode); work assistant (TraeWork); web (TraeWork Web)	Official
OS	Desktop IDE + browser (web edition)	Official
Browser / computer use	Yes (TraeWork Web)	Vendor-reported
Coding	Yes (TraeCode)	Vendor-reported
Autonomous task execution	Yes (coding agent, work assistant)	Vendor-reported
MCP	Not publicly documented	Not publicly documented
Tool calling	Yes (coding tools)	Vendor-reported
Memory	Not publicly documented	Not publicly documented
Workflow	Coding agent + work assistant under one product line	Vendor-reported
API	Not publicly documented	Not publicly documented
Pricing	Not publicly documented on the marketing page	Not publicly documented
Region	Global	Official
Open-source	No — proprietary	Official
Deployment	Cloud	Official
Limitations	Underlying model unnamed; thin public docs; closed product	Official
Source	trae.ai · TraeWork Web	Official
Last verified	2026-09-30	Official

See the ByteDance company profile, the choosing-a-coding-model guide, the choosing-an-agent guide, and the site’s AI agents and tool calling technology pages. For the structural reading of Trae against the coding specialists, see the research on the agent-led coding race and Chinese AI coding models.

Labels used above: Official fact (from the Trae site, TraeWork Web edition and TraeCode IDE page), Vendor-reported claim (positioning statements by ByteDance), and China AI Hub analysis (our synthesis, always introduced as such). No third-party evaluation evidence is currently recorded for Trae.