Devin Desktop (Formerly Windsurf): What Changed

6 min read

TL;DR

• Devin Desktop dropped prompt credits in March 2026. Your plan now gives you a daily and a weekly usage allowance that refreshes on its own (quota docs).
• How fast you burn that allowance depends on the model’s token cost. Free models don’t count against it at all.
• When you run out, Pro, Teams and Max users can buy extra usage, billed at API list prices. Free users wait for the next reset.
• If you were already paying $15/mo for Pro or $30/seat for Teams, you keep that price indefinitely.
• SWE-2, Cognition’s own coding model, is free on self-serve plans only through October 15, 2026. After that it costs $3.00 per million input tokens and $15.00 per million output tokens (models docs).

Remember when Windsurf was the cool, simple one? Yeah. Then it became Devin Desktop, and the credit system got the full Ship of Theseus treatment: every plank replaced, the name on the side changed, and now you’re not quite sure it’s the same boat. Here’s what actually changed, and the parts that’ll hit your wallet.

The short version: credits are gone, budgets are in

The old deal was simple to read and hard to plan around: each prompt cost credits, and you bought more when you ran dry. Devin Desktop replaced that in March 2026 with a quota-based system. Your plan includes a daily and weekly budget, and each request draws from it according to how many tokens the model uses (quota docs).

In practice, a quick prompt with a couple of files open costs little, and a long agent session over a big codebase costs a lot. The docs say this “better reflects the underlying costs of using different models,” which is corporate for “you power users were having way too much fun.”

One genuinely nice touch, credit where it’s due (sorry, couldn’t resist), if you code on weekends. Your daily quota is set higher than one-seventh of your weekly quota, so you can do a big Saturday session without being cut off by the daily cap. Quotas reset by calendar date.

What happens when you hit the limit

This depends on your plan:

Plan	When the quota runs out
Free	Wait for the next daily or weekly reset
Pro, Teams, Max	Buy extra usage and keep working

Source: Devin Desktop quota docs, checked Oct 14, 2026.

Extra usage is billed at API list prices for whichever model you’re using. Faster or priority variants, like SWE-1.5 Fast or the fast Opus variants, cost more. Quota caps never limit extra usage. They only cap the allowance that comes with your plan.

Model choice is now a price decision

Under credits, picking a model mostly decided quality. Now it also decides how long your budget lasts. Here’s the self-serve pricing the docs list right now:

Model	Input (per 1M tokens)	Output (per 1M tokens)	Note
SWE-2	Free through Oct 15, 2026, then $3.00	Free, then $15.00	Cache input $0.30 after the free period
Gemini 3.7 Flash / 3.8 Flash	$0.75 (intro)	$3.75 (intro)	Rises to $1.50 / $7.50 after Dec 31, 2026
Gemini 4 Argon	$1.00 (extra 50% off through Oct 27)	$5.00	Intro $2.00 / $10.00 through Dec 1, then $4.00 / $20.00

Source: Devin Desktop models docs, checked Oct 14, 2026. The docs point to the in-app model selector for the most current prices.

The docs suggest Fusion for most users. It pairs a frontier lead model with a cheaper sidekick model to keep costs down. Cognition’s in-house SWE family covers the rest:

• SWE-2: the newest, in Medium, High and Max reasoning effort.
• SWE-1.7: the previous flagship. A Lightning version runs on Cerebras for lower latency.
• SWE-1.6: the previous generation. A Fast version is available to paying users.
• SWE-1-mini, swe-grep and swe-check: these power Tab suggestions, context retrieval and Quick Review.

The practical move: use SWE-1.7 or SWE-1.6 for routine edits, as the docs themselves recommend, and save frontier models for hard problems. Sticking with one frontier model per task helps too, because repeat requests to the same model reuse its cache and use fewer tokens.

Billing extra usage at full API list prices, with faster variants costing more on top, means heavy agent users now pay retail for the exact workflow the product is built around. That one stings.

What 200,000 tokens looks like

The docs walk through one refactoring conversation with Claude Opus 4.6. The prompt is “Refactor @my_function”, the agent analyzes it, proposes a plan, gets a “yes, continue,” then edits one file and adds another. That one task adds up to about 200k tokens, because every step re-sends the shared context: the timeline, the editor state and the system prompt (quota docs).

That’s the main thing to understand about the new system. A short chat back and forth isn’t what drains your quota. Agent loops that keep re-reading a growing context are. Your agent is basically Dory from Finding Nemo, re-reading the whole conversation every single step, and you’re paying for every re-read.

If you were already a subscriber

• Pro at $15/mo: grandfathered at $15/mo indefinitely, on the new quota system.
• Teams at $30/seat/mo: grandfathered at $30 per Developer seat indefinitely.
• Free week: every existing paid subscriber got 7 extra days added to their renewal date, to try quotas before committing.
• Annual plans: the renewal date also moved by 7 days. If you cancel, you can request a refund for the remaining months.
• Unused add-on credits: converted to extra-usage dollars at the rate you paid, $0.04 per credit, so 250 credits became $10. You can ask support to refund that balance at any time.
• SSO: if you bought the SSO add-on, you keep it on your grandfathered plan. New Teams plans no longer include SSO, since it’s now Enterprise-only.
• Enterprise customers: existing billing agreements stay in place for now.

Source: Devin Desktop quota docs.

Is it still worth it?

If you’re grandfathered at $15, yes, probably. Hold onto that plan like it’s the last slice of pizza. You keep a low price and get a free SWE family that the docs push for everyday work. Just watch the usage meter in the app during your first heavy week.

If you’re new, check current plan prices on Devin’s site, then compare usage limits rather than sticker prices. Most AI coding tools sell an entry tier and a pricier power tier, and the real difference is how far each one stretches. See our AI coding tools comparison and the pricing tracker.

If you live in agent mode on frontier models, budget for extra usage at API rates. That’s where the new system costs more than credits did.

FAQ

Did Devin Desktop get rid of credits? Yes. Since March 2026, plans use daily and weekly quotas instead of prompt credits.

Do free models use up my quota? No. The docs say free models don’t count against your quota.

What happens to my old add-on credits? They were converted to extra-usage dollars at the rate you paid ($0.04 per credit), and you can request a refund of that balance.

Is SWE-2 free? On self-serve plans it’s free through October 15, 2026. After that, list pricing applies.

Sources

• Devin Desktop: Quota-Based Usage, checked Oct 14, 2026
• Devin Desktop: AI Models, checked Oct 14, 2026