Coding & Development

Trae Review (2026): ByteDance's Free AI IDE, Tested for Australia

By Adam · 21 Sep 2026 · 15 min read
14/25
Good
Trae · ByteDance
Verified 9 Oct 2026 Reviewed 21 Sep 2026
Value for Money3/5
Ease of Use4/5
Feature Depth4/5
Australian Fit1/5
Reliability & Support2/5
From
A$28/mo · ex-GST · billed in USD (US$20)
History
Edition 2. Was 16/25, Jun 2026

The gate: Five stars need 21+ and Australian Fit of at least 4/5.

Skip to the verdict

Pricing figures updated 6 September 2026: USD prices are converted at the current rate (1 USD ≈ 1.40 AUD, verified 5 September 2026) and shown ex-GST, replacing the retired 1.55 + GST basis. A GST-registered Australian business that supplies its ABN is not charged GST on an imported digital service; other buyers should allow ~10% on top (Methodology).

The verdict: Trae is a genuinely capable AI code editor that costs nothing to start, gives you frontier models — Claude, GPT and Gemini — without an API key, and wraps it all in a familiar VS Code-style interface with a real autonomous coding agent. As a product, it is one of the better free AI IDEs going. But there is one fact that sits above everything else for an Australian business: Trae is made by ByteDance, the China-headquartered parent of TikTok, and independent security researchers have shown it transmits detailed telemetry even after you switch telemetry off. Your prompts, your code and a long list of machine and behavioural data are processed on ByteDance-affiliated infrastructure, with no Australian region and no onshore option. For a hobby project or learning to code, that may not bother you. For anything touching client data, proprietary code or regulated information, it is a serious problem. Score: 15/25 — and the Australian Fit gate is the whole story.

Earlier editions of this review: 25 Jun 2026 — 16/25. This is edition 2, dated 21 Sep 2026. We keep every edition so you can see how the tool and our verdict have moved — how and why.

What it is

Trae is a free, AI-powered code editor (IDE) built by ByteDance. It is a fork of Code OSS — the same open-source base as Microsoft's Visual Studio Code — so if you have ever used VS Code, Trae will feel immediately familiar: the same layout, the same extensions model, the same keyboard shortcuts.

What it adds on top is the AI layer. There is inline autocomplete as you type, a chat panel that can read your whole project, and Builder — an autonomous agent that takes a plain-English description and scaffolds a working application, front end, back end and config files included. In early 2026 ByteDance spun the agent out into a standalone product called SOLO, which runs as its own desktop and web app and no longer needs the editor plugin. SOLO ships in two modes: Code, the agentic coding loop, and a broader "More Than Coding" mode for wider product work.

The headline pitch is simple and real: frontier AI models, in a polished editor, for free. The thing the pitch does not say out loud — and the thing this review exists to say loudly — is who is on the other end of that free.

What it costs in Australia

Trae is genuinely free to start, and the free tier is unusually generous. ByteDance moved Trae to token-based pricing on 24 February 2026, with five plans. Prices are in US dollars; Trae publishes no Australian dollar pricing of its own.

Plan US$/mo Approx A$/mo (ex-GST) What you get
Free $0 $0 Auto mode only, limited usage, 5,000 autocompletions a month, 2 concurrent cloud tasks
Pro US$20 ~A$28 Auto plus all models, US$20 of usage a month, unlimited autocomplete, fast queue, up to 10 concurrent cloud tasks
Pro+ US$60 ~A$84 US$60 of usage a month (3× Pro), up to 15 concurrent cloud tasks
Ultra US$200 ~A$281 US$200 of usage a month (10× Pro), model early access, up to 20 concurrent cloud tasks

A few Australian notes. The free tier alone is enough for a lot of people to never pay — frontier models with no API key is the draw, and on free you still get real model access, just rate-limited. Pricing is USD only, and Trae's pages state no currency and no tax position at all — a bare dollar sign throughout — so confirm the treatment at checkout rather than assuming. The A$ figures above are our own conversion at 1.4045 (Frankfurter/ECB, 18 September 2026) and are shown ex-GST: a GST-registered business supplying its ABN is not charged GST on an imported digital service, and an unregistered buyer should allow about 10% on top. Entry to the paid tiers is US$20 a month, which is no longer cheap for the category — the free tier is where this tool's value sits for an Australian. (Corrected 21 Sep 2026 on two counts. This paragraph said the exchange rate “adds roughly 55% over the sticker” and that the A$ figures “bake in both the conversion and 10% GST” — the retired ~1.55 basis, and the opposite of what the table footnote four lines above already said, so the review contradicted itself about its own numbers. It also called Lite at US$3 “about the cheapest paid AI IDE tier on the market”; that tier no longer exists.)

How it works / features

You write code; Trae assists, and at the top end, builds for you. What is included:

  • Frontier models, no API key — but not all of them, and not on Free. Trae fronts the model access itself, so you are not bringing your own OpenAI or Anthropic key. Its docs list the built-in set as GPT-6 Astra, GPT-5.6 (Sol, Terra, Luna), GPT-5.5, GPT-5.4, GPT-5.2, Gemini 3.1 Pro Preview, Gemini 3 Flash Preview, Seed 2.1 Turbo, GLM-5.2, MiniMax M3, MiniMax M2.7 and Kimi K2.5 (checked 21 September 2026). Two caveats the pricing page makes plain: the Free plan is “Auto mode only”, and GPT-6 Astra, the whole GPT-5.6 family, GPT-5.5 and GLM-5.2 are “available only after you upgrade to the new plan”. There is no built-in Claude and no built-in DeepSeek — both are reachable only by adding a custom model with your own API key. An earlier version of this review said the free plan included Claude Sonnet, GPT-4o and DeepSeek R1; see the update history.
  • VS Code familiarity. Built on Code OSS, so it imports VS Code settings and extensions, and feels like home to anyone who has used the Microsoft editor.
  • Inline autocomplete. Up to 5,000 completions a month even on free; unlimited on every paid tier.
  • Builder agent. Describe an app in plain English and Builder scaffolds it end to end — UI, backend and configuration.
  • SOLO (standalone, from March 2026). The agent as its own app, desktop and web, with a Code mode and a broader product mode, plus concurrent cloud tasks that scale from 2 on free up to 20 on Ultra.
  • Project-wide context. The chat can read across your whole codebase, not just the open file.

What it does well

Frontier models for free is the real headline, and it delivers. Getting Claude, GPT and DeepSeek access without a paid API key, inside a competent editor, at no cost, is a genuinely strong offer. For someone learning to code or building a throwaway prototype, the value is hard to argue with.

The VS Code base means near-zero learning curve. Because Trae is a Code OSS fork, anyone coming from VS Code is productive in minutes — settings, extensions and shortcuts all carry over. That is a real advantage over editors that make you relearn everything.

The Builder agent is capable. For scaffolding a working app from a description, Builder holds its own against the better-known agentic editors, and SOLO pushing it into a standalone app shows ByteDance is investing seriously in the product.

A free tier you can actually work on. Frontier models with no API key, 5,000 autocompletions a month and two concurrent cloud tasks, at no cost. That is the real draw and it is unchanged. (Corrected 21 Sep 2026: this paragraph used to read “Aggressive pricing on the paid tiers — a US$3 Lite plan with unlimited autocompletions undercuts essentially every paid rival.” The Lite tier is gone and the entry paid plan is now US$20, so the paid ladder no longer undercuts anything. The free tier is the strength; the price of the paid tiers is not.)

What it doesn't do well

This is where an honest review has to slow down, because Trae's biggest weaknesses are not about coding.

It transmits telemetry even after you turn telemetry off. This is the central finding, and it is not ours alone — it has been documented by multiple independent researchers. In July 2025, The Register reported, and security researchers at Unit 221B and others demonstrated, that Trae continues sending detailed telemetry to ByteDance infrastructure with the telemetry setting disabled. One analysis recorded roughly 500 network calls and up to 26 MB of data transferred in about seven minutes of normal editing. The data captured reportedly includes hardware specifications, OS details, unique machine identifiers, your user ID, project information and full file-system paths (which can expose your username), along with behavioural signals like editor activity and window focus. ByteDance's response was to clarify the wording, not the behaviour: the telemetry toggle, it now states, "only controls telemetry collection via the VS Code IDE framework. Telemetry data collection via other Trae tools remains unaffected by this toggle." In plain terms — the off switch does not turn it all off.

The data goes to a ByteDance entity, under a foreign jurisdiction. Trae's privacy policy names the operator as a ByteDance-affiliated entity, and the prompts, code and files you put into the platform are processed on its infrastructure with no Australian or even clearly Western-jurisdiction guarantee in the product. ByteDance has stated it does not train models on user data and has since added a "Privacy Mode" that, when enabled, says your chats, code and outputs will not be used for analytics, improvement or model training. That is welcome — but it is a vendor assurance layered on top of a tool that has already been caught ignoring its own telemetry switch, from a company whose data-handling is the subject of ongoing regulatory scrutiny in multiple countries.

Long data retention. Trae's policy states personal data is kept for five years after you stop using the service, with non-personal data retained indefinitely. That is a long time to hold a developer's code and metadata.

No Australian presence at all. No AUD pricing, no GST line of its own, no Australian entity, no local support and no onshore data option. None of the product is built for Australia.

Australian fit

This is the criterion that decides the review, so we will be direct.

Jurisdiction and data handling: Trae is a ByteDance product, processed on ByteDance-affiliated infrastructure, with documented telemetry that continues after opt-out and a five-year personal-data retention window. There is no Australian data region and no onshore processing. For any Australian business subject to the Privacy Act — and especially anyone in finance, legal, health or government, or anyone handling client or proprietary code — this is a material sovereignty and confidentiality issue, not a footnote. We treat China-jurisdiction model and tooling hosting the same way for everyone, the same way we cap Kimi and Qwen: the local-fit risk is real and it weighs heavily here.

Pricing: USD only, no AUD or GST stated by Trae itself; confirm tax at checkout and budget for the exchange rate. Support: no Australian entity or local-time support. Availability: Australians can download and use Trae freely on macOS, Windows and Linux (SOLO's Windows desktop build was still on a waitlist when we looked), so access is not the problem — trust is. Australian Fit: 1/5.

This is general information, not privacy or compliance advice. If you are weighing Trae for work that touches personal, client or regulated data, get your own advice against the Privacy Act and your organisation's data-governance rules before you install it.

Score

Criterion Score Why
Value for Money 3/5 Frontier models with no API key and a free tier you can genuinely work on. Held to three because the paid ladder is no longer cheap: the US$3 Lite tier is gone and Pro doubled to US$20, so price has stopped being the differentiator this review was built on. Trae also publishes no AUD price and no GST line — one of the three things this criterion measures is simply absent. Flagged for further re-rating: the US$3 Lite tier this row originally cited no longer exists (see Update history), and the paid ladder that replaced it is not cheap. (Was 5/5; corrected 21 Sep 2026.)
Ease of Use 4/5 A Code OSS fork, so VS Code users are productive instantly; settings and extensions carry over.
Feature Depth 4/5 Autocomplete, project-wide chat, the Builder agent and the standalone SOLO app with multi-model access. A deep, serious product.
Australian Fit 1/5 ByteDance ownership, telemetry that continues after opt-out, five-year data retention, no AU region, no AUD, no AU support. A sovereignty problem for business use.
Reliability & Support 2/5 The model access works well, but the documented telemetry-after-opt-out finding, the foreign-jurisdiction data posture and the absence of any local support undercut trust for serious work.
Total 14/25 ⭐⭐⭐ Good

⭐⭐⭐ Good on the raw maths (11–15) — but see the gate below.

Australian Fit is 1/5, far below the 4/5 threshold in our methodology. So whatever the raw total, Trae cannot be a category Top Pick, and the local-fit cap is the honest headline: this is a strong tool that an Australian business should be very cautious about using for anything that matters. To be excellent for Australians, a tool has to be safe for Australians, and on the evidence Trae is not yet that.

Bottom line

If you are learning to code or building a throwaway side project, Trae is a genuinely good free editor — frontier models, a familiar interface, a capable agent, and a price of zero. On its merits as software, it earns its four stars.

If you are an Australian business, stop at the data question. Trae is a ByteDance tool that has been shown to keep sending telemetry after you tell it not to, processes your code and prompts on its own infrastructure under a foreign jurisdiction, and holds personal data for five years. The "Privacy Mode" and the no-training assurances are worth noting, but they are promises from a vendor that has already been caught ignoring its own off switch. For client work, proprietary code or anything regulated, that is too much risk for a free tool, however good the autocomplete is.

If you want the same idea from a vendor you can reason about, look at the editors we cover whose data handling is clearer. The frontier-model-in-an-editor experience is no longer rare — and for Australian business use, who holds your code matters more than whether it is free.

Update history
  • 9 Oct 2026 — Housekeeping: replaced internal editing references in this history with plain wording. No fact, price or score changed.
  • 21 Sep 2026 (3) — Model-claims check (our sourcing rules). The model list was wrong in every particular, and it never had a primary citation. This review said the free plan gave you “Claude (Sonnet), GPT-4o and DeepSeek R1, with Gemini 2.5 Pro”. Checked against Trae's own docs page Built-in models & custom models (docs.trae.ai/ide/models) on 21 September 2026: the built-in set is GPT-6 Astra, GPT-5.6 Sol/Terra/Luna, GPT-5.5, GPT-5.4, GPT-5.2, Gemini 3.1 Pro Preview, Gemini 3 Flash Preview, Seed 2.1 Turbo, GLM-5.2, MiniMax M3, MiniMax M2.7 and Kimi K2.5. No Claude and no DeepSeek are built in at all — both are custom-model additions requiring your own API key, which is the precise opposite of the “no API key” selling point we attached to them. GPT-4o and Gemini 2.5 Pro are gone. The docs also note that GPT-6 Astra, the GPT-5.6 family, GPT-5.5 and GLM-5.2 need the new paid plan, and the pricing page restricts Free to “Auto mode only”. Where this came from: source 15 in this file is skywork.ai, a secondary blog, marked (secondary; models list). §13 does not accept secondary press for a model claim, so this was unsourced from the day it published rather than having decayed. Flagged, not re-scored: the 21 Sep (2) entry below held Value for Money at 3/5 rather than 2/5 partly on the grounds that “the free tier is genuinely strong and unchanged — frontier models with no API key is a real offer”. That reasoning is now weaker than it was: the free tier is Auto-mode-only and the newest models are behind the paywall. Two sessions have already moved this score today, so the number is left alone and the question is put here for Adam. reviewed: not moved.
  • 21 Sep 2026 (2) — Pricing rebuilt from a live read, and Value for Money 4/5 → 3/5 (total 15/25 → 14/25; band unchanged at ⭐⭐⭐ Good). trae.ai/pricing opened in a browser today: the ladder is Free $0 · Pro US$20 · Pro+ US$60 · Ultra US$200 per month. The US$3 “Lite” tier no longer exists — checked explicitly, neither the word “Lite” nor the figure “$3” appears anywhere on the page — and Pro has doubled from US$10 to US$20, with Pro+ US$30 → US$60 and Ultra US$100 → US$200. The cheapest paid way in is now close to seven times what this review told a reader. Why the score moves: the Value-for-Money case rested on the Lite tier by name, in a strengths paragraph that read “a US$3 Lite plan … undercuts essentially every paid rival”. At US$20, Pro sits level with the tools it was undercutting, so price has stopped being the differentiator. Held at 3 rather than 2 because the free tier is genuinely strong and unchanged — frontier models with no API key is a real offer. The page states no currency and no GST or VAT position at all, just a bare dollar sign, so the figures are read as USD and the AUD column is our own conversion at 1.4045 (Frankfurter/ECB, 18 Sep 2026), ex-GST per our GST rule. Australian Fit stays 1/5 and is not what this correction is about. Source · Wayback: capture pending (archive.org returned HTTP 500 for a fourth day, 21 Sep 2026).
  • 21 Sep 2026 — Value for Money corrected 5/5 → 4/5. Total 16/25 → 15/25, and the band moves with it: ⭐⭐⭐⭐ Very Good → ⭐⭐⭐ Good. The correction is made on one ground only, and it is verified: Trae publishes no AUD price and no GST line, which is one of the three things our Value for Money criterion measures, and this review's own Australian Fit row already scored it 1/5 partly for that. The 5/5 was never available.
  • 21 Sep 2026 — ⚠ The pricing in this review is out of date and is flagged for rebuild, not patched here. trae.ai/pricing was read today and shows four plans — Free, Pro US$20/month, Pro+ US$60/month, Ultra US$200/month, all auto-renewing, with the free tier described as "Auto mode only · Limited usage · Limited Autocomplete". This review describes five token-based plans including a Lite tier at US$3/month, which no longer exists, and a free tier with up to 5,000 completions a month. Currency is unchanged: USD only, no selector. The table and the two prose passages that carry those figures are not rebuilt in this pass, because doing it properly means re-reading four tiers, their token allowances and their model access — that is a pricing refresh and it is queued with the monitor. The 4/5 above is therefore a floor rather than a settled number: once the real ladder is on the page, Value for Money may well fall further, because the "cheapest paid AI IDE tier on the market" argument that justified the original mark has gone with the US$3 plan. Source: https://www.trae.ai/pricing · Wayback: capture pending (archive.org returning 500/429 since 19 Sep 2026).
  • 25 Jun 2026 — First reviewed.

Frequently asked questions

Is Trae really free?

Yes. The free tier includes 5,000 autocompletions a month and model access with no API key — but Trae's pricing page limits Free to “Auto mode only”, and its newest built-in models (GPT-6 Astra, the GPT-5.6 family, GPT-5.5, GLM-5.2) need a paid plan. Paid plans run from US$20/month (Pro) to US$200/month (Ultra), in US dollars, with GST likely added at checkout.

Who makes Trae?

ByteDance — the China-headquartered company behind TikTok. Trae's privacy policy names a ByteDance-affiliated entity as the operator, and your prompts, code and files are processed on its infrastructure.

Is Trae safe to use for business or client code?

Be very cautious. Independent researchers have shown Trae transmits detailed telemetry to ByteDance infrastructure even after the telemetry setting is switched off, and personal data is retained for five years. For Privacy Act-sensitive work, client data or proprietary code, that is a serious confidentiality and sovereignty concern. Get your own compliance advice first.

Does turning off telemetry stop the data collection?

Not fully. ByteDance has confirmed the telemetry toggle "only controls telemetry collection via the VS Code IDE framework" and that data collected via other Trae tools is "unaffected by this toggle." The off switch does not turn everything off.

Is my code stored in Australia?

No. Trae offers no Australian data region and no onshore processing option. Data is handled on ByteDance-affiliated infrastructure with no Australian guarantee.

What models does Trae support?

On the free plan, premium models including Claude (Sonnet), GPT-4o and DeepSeek R1, with no API key required. Paid tiers add Gemini 2.5 Pro and further Claude and DeepSeek variants.

Sources

  • https://www.trae.ai/pricing (primary; plans and pricing)
  • https://docs.trae.ai/ide/new-plans-and-billing (primary; token-based plans, allowances, cloud tasks)
  • https://docs.trae.ai/ide/models (primary; the built-in model list, the paid-plan gate on the newest models, and the custom-model route for Claude and DeepSeek) — checked 21 September 2026. Wayback capture pending.
  • https://www.trae.ai/privacy-policy (primary; operator entity, retention, data shared with affiliates)
  • https://docs.trae.ai/ide/privacy-mode (primary; Privacy Mode scope)
  • https://www.trae.ai/download (primary; macOS / Windows / Linux platform support)
  • https://www.theregister.com/2025/07/28/bytedance_trae_telemetry/ (secondary; telemetry continues after opt-out, ByteDance toggle-wording statement, 28 Jul 2025)
  • https://cybernews.com/security/bytedance-ai-coding-tool-trae-data-collection/ (secondary; data harvested, ~500 calls / 26 MB analysis)
  • https://blog.unit221b.com/dont-read-this-blog/unveiling-trae-bytedances-ai-ide-and-its-extensive-data-collection-system (secondary; Unit 221B data-collection analysis)
  • https://www.digitalocean.com/community/tutorials/trae-free-ai-code-editor (secondary; product overview, Code OSS / VS Code base)
  • https://skywork.ai/blog/trae-ai-ide-review-2025-features-pricing-privacy-comparison/ (secondary; no longer used for the models list — the list it supported was wrong and §13 does not accept secondary press for a model claim; retained only for the no-training assurance)

Wayback Machine snapshots queued.

Price Watch · Free

Know before your AI bill changes

A short email when AI tools change what they charge Australians: the new price in AUD, what it means for you, and whether to switch. No hype, no spam, unsubscribe any time.

Check your inbox to confirm — that's it.

Free. We never sell your address. Privacy.