Trae by ByteDance Pricing: What You Actually Pay in 2026

ByteDance’s AI IDE switched to token-based billing in February 2026, replacing the informal free-forever model that made early adopters ditch Cursor. There are now four tiers—Free, Lite ($3), Pro ($10), Ultra ($100)—and the free tier still exists, but it’s not a free pass anymore.

What Changed in February 2026

Before February, Trae offered unlimited access to Claude Sonnet and GPT-5.4 with no tier wall. ByteDance subsidized the entire operation to grow market share against Cursor’s established user base. That calculus shifted. The new token-consumption model means you now pay for what you use—or you don’t use Trae much at all.

The move mirrors what happened in the broader AI IDE space: free tiers exist to onboard, but heavy usage demands paid plans. For developers who were riding the free tier, this was the moment the math changed. For those already on Lite or Pro, the transition was smoother—the subsidy just got privatized into your subscription.

The Four Tiers: Free, Lite, Pro, Ultra

The pricing table is straightforward:

The Lite tier comes with a WELCOME25 promo code—first month is $3 as an introductory rate, though the messaging on whether that price sticks is ambiguous. Pro can be paid annually at $90/year (equivalent to $7.50/month if you commit upfront), which shaves 25% off monthly cost. Ultra pricing is flat and not discounted.

The Free Tier Is Still Real — But It Has a Ceiling

We expected Trae to pull the free tier entirely. They didn’t. But “free” now carries a hard asterisk: rate limiting kicks in during peak hours. Your requests queue instead of executing immediately. For a casual developer checking Trae once or twice a day, this is fine. For someone building a feature and iterating rapidly, it’s friction.

The trade-off is obvious: zero dollars, or pay $10 to remove the queue and get 600+ guaranteed fast requests per month. If you’re testing whether Trae fits your workflow, free still lets you kick the tires. If you’re actually building with it, the free tier will frustrate you into the Pro bucket within a week.

Is the $3 Lite Plan Worth It?

The Lite tier is a positioning play. At $3 per month, it’s entry-cost furniture—low enough that it feels free, high enough that you’re a paying customer. The plan includes $5 in monthly usage credit, which for light Builder Mode work might cover a week or two of genning boilerplate and refactoring.

Here’s where we’ll stake a take: Lite is a trap. You’ll hit the usage ceiling mid-session, get frustrated, and either walk back to free (and suffer the peak-hour queue), or jump to Pro. It’s the classic “mid-tier designed to funnel you upward” move. Trae isn’t hiding that—the Lite page doesn’t pretend you’re getting a full development experience. It’s honest about the cap. But it’s still a cap, and $3/month psychology means you’re more likely to upgrade than if the tier didn’t exist.

That said, if you’re primarily using Trae for code review and light refactoring, Lite can work. The key is knowing your own usage pattern before you commit. Don’t sign up for Lite expecting to do heavy Builder Mode sessions without hitting the token wall.

Pro at $10: Where the Math Gets Interesting

Pro is where Trae becomes a real Cursor alternative. At $10/month (or $90/year), you get 600+ fast premium requests, priority support, and unthrottled access to Claude Sonnet 4 and GPT-5.4. Cursor’s standard plan is $20/month for comparable model access, which makes Trae’s Lite + Pro combo an 85% savings play if you’re coming from Cursor’s baseline.

We tested this math ourselves: if you’re paying $20 for Cursor and using it for everyday refactoring and feature planning, Trae Pro at $10 delivers the same core experience at half the cost. The UI is different, the model latency varies slightly, but the capability floor is equivalent. 600 fast requests per month is roughly 20 per day—enough for serious development work without the paranoia of running out.

The annual prepay ($90/year instead of $120 if billed monthly) sweetens it further for committed users. That’s a real discount, not a fake “save $30 over 12 months” marketing number. Lock in $7.50/month, get the full Pro suite, and you’ve neutralized most of Trae’s pricing objections versus Cursor.

Ultra at $100: Who Is This For?

Ultra is the enterprise tier. $100/month gets you maximum token pool, early access to new models, and priority everything. This tier exists for teams, not solo developers. If you’re building a product on top of Trae’s API or running a dev shop with multiple seats, Ultra’s early model access and guaranteed throughput matter. For an individual developer, it’s overkill unless you’re running continuous code generation at scale.

We’d only recommend Ultra if you’re hitting Pro’s limits regularly and your work depends on the absolute latest model releases. Otherwise, Pro covers 90% of individual developer workflows.

Our Verdict on Trae Pricing

ByteDance built a pricing model that undercuts Cursor while staying sustainable. The free tier is real for evaluation. Lite is a honeypot. Pro is the actual value tier. Ultra is for teams.

If you should pay for Trae: You’re a solo developer or small team currently on Cursor or another paid IDE, and you want to cut your subscription cost in half while getting equivalent Claude and GPT access. Switch to Pro at $10/month, trial it for a month, and decide if the interface and latency work for you.

If you shouldn’t: You’re committed to Cursor’s UX, or you need features Trae doesn’t yet support (like native remote development). The price cut isn’t worth a workflow retrain.

The Feb 2026 billing shift removes the “just use it free forever” equation that made Trae a no-brainer for early adopters. Now it’s a normal pricing decision: does Trae’s product at $10/month beat Cursor at $20/month for your use case? If yes, the math wins. If the interface feels foreign or the latency stings, no discount is worth the friction.

For other IDE pricing moves in this window, see how Cursor’s own four-tier structure and Windsurf’s quota flip compare.