Disclosure: some links on this site may be affiliate links. If you buy through one, we may earn a commission at no extra cost to you. It never changes what we write or how we rank things. Details.

Windsurf (Devin Desktop) quotas explained: daily and weekly limits, extra usage and token prices (checked November 2026)

Updated November 14, 2026 · Facts come only from the vendors' official pages listed under Sources. We have not run paid accounts on these products for this page.

Windsurf is now called Devin Desktop, and since March 2026 its plans no longer come with a monthly pile of prompt credits. You get a daily and weekly quota measured in tokens, and anything past it is billed as extra usage at API list prices. The pricing page tells you the plan prices but not how big the quotas are. Below is what Devin's own pricing page and docs said on November 14, 2026, and a worked example of what an agent session costs once you are past the quota.

The short answer

• Plans: Free ($0), Pro ($20/month), Max ($200/month) and Teams ($40/month per full seat with an $80/month minimum, see below). Enterprise is billed in Agent Compute Units (ACUs) under a contract.
• The quota is a daily and weekly budget of tokens. Each model has its own per-token cost, so a frontier model uses it up faster than a small one, and "free models don't count against your quota at all." Max has a weekly quota only, with no daily cap.
• Quota sizes are not published. The pricing page says only that Pro has "Increased quotas" and Max "Significantly higher quotas", and that "For the majority of users, this quota will be enough."
• Past the quota, Free users wait for the reset. Pro, Max and Teams can keep going on extra usage, prepaid and "billed at API list prices for the model you're using." The self-serve docs call this balance on-demand credits; purchased credits roll over and never expire.
• Two in-house models are free for now: SWE-2 in Devin Desktop and Devin CLI through November 30, 2026, and SWE-2 Flash until December 4, 2026 at 18:00 UTC.

What changed in March 2026

Devin's docs say that in March 2026 Devin Desktop "replaced the credit-based system with a quota-based usage system." Before that, Windsurf plans came with built-in prompt credits, and add-on credits were sold at $0.04 each.

• Existing subscribers keep their old price: Pro at $15/month and Teams at $30 per Developer seat, "grandfathered in … indefinitely", but on the new quota system.
• Leftover add-on credits were converted to extra-usage balance at the rate paid: every 250 credits became $10. The docs say you can ask support for a refund of that balance at any time.
• Enterprise customers stay on their existing billing agreements. Some older Enterprise contracts still use prompt credits; newer ones use ACUs.

What each plan includes

Notes on the table:

• Resets follow the calendar date. On Pro and on Teams full seats, "the daily allowance is more than 1/7 of the weekly", so someone who also works weekends can still use the whole weekly allowance.
• Pro and Max are single-user plans. To share one subscription between people you need Teams.
• No free trial of paid plans. The docs say you can request a refund before you start using a subscription.
• Where to see what's left: the usage meter in Devin Desktop (status bar widget, "Plan Info"), or Settings > Usage & limits on app.devin.ai.

Teams: what you actually pay

Teams has two kinds of seat. A full seat costs $40/month and comes with its own Pro-sized quota. A flex seat is free, has no quota of its own and draws everything from the team's shared on-demand credits.

Devin's self-serve docs say every Teams account pays at least $80/month, and full seats count toward that minimum. Anything under $80 is charged as prepaid on-demand credits the team can use:

The pricing page's headline reads "$80 /month for team plan + $40/mo per full dev seat", which could be read as $80 on top of the seats ($280 for five full seats). We go by the docs, which spell the bill out row by row, but check the total at checkout. On Teams, Automations and Devin Review also draw from the shared credits, not from full-seat quota.

How the quota is used up

The docs describe the mechanics briefly: the budget "is based on how many tokens the model uses for each request"; "the cost per token varies by model"; short requests with a few files in context use fewer tokens than long requests over a large codebase; and free models don't count.

The quota and the per-token prices are not the same thing. The models page says its table "shows token rates for extra usage. Quota consumption can differ from these rates." So the prices below tell you what extra usage costs in dollars, but not what share of your quota a request takes.

Devin's quota page includes an example: one agent conversation with Claude Opus 4.6 that adds up to 200,000 tokens over two user messages, three tool calls and three replies. Most of those tokens are context that is re-sent on each call and read from the cache. The example doesn't split each row into cache reads and cache writes, so we haven't priced it; the point is that a short back-and-forth with an agent already runs to hundreds of thousands of tokens.

Extra usage: per-token prices

These are the self-serve rates from Devin's models page (the "Recommended" view, Medium reasoning unless noted), which the page dates "as of November 9, 2026". The full table also lists other effort levels and older models.

All prices are US dollars per million tokens.

Dated changes the page lists:

• SWE-2: the table itself shows dashes for SWE-2; the note above it says SWE-2 is free for self-serve plans in Devin Desktop and Devin CLI through November 30, 2026 (not in Devin cloud sessions). After that, list pricing applies: $3.00 input, $15.00 output and $0.30 cached input per million.
• SWE-2 Flash: free until December 4, 2026 at 18:00 UTC, then $0.10 input, $0.01 cache read, $0.125 cache write and $0.50 output per million. Free plans can use the Medium variant only.
• Gemini 4 Argon: the prices above run until December 1, 2026 at 08:00 UTC, then $4.00 input, $20.00 output and $0.20 cache read per million.
• Gemini 3.8 Flash: introductory pricing through December 31, 2026, then $1.50 input and $7.50 output per million.
• Fast and priority variants cost more. For example, the "Fast" variants of Claude Opus 5.5 and Claude Fable 5.2 are listed at twice the standard rates.
• Long context: for GPT-6 Astra, Luna and Sol, a request whose total prompt (uncached input, cache reads and cache writes) goes above 272K tokens is billed entirely at higher long-context rates.

For the models both vendors list (GPT-6 Luna, Gemini 3.8 Flash, Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.2 and GPT-6 Astra), these input, cache-read and output rates are the same as those in GitHub's table for Copilot AI credits, which we covered in GitHub Copilot AI credits explained.

Worked example: one agent session past the quota

We use the same made-up session as in our Copilot article: 30 model calls, each sending 40,000 input tokens and getting 1,500 back, so 1.2 million input tokens and 45,000 output tokens. We show it with nothing cached and with 90% of each call's input read from the cache. Cache writes are left out, so treat the cached column as a floor.

How to read it:

• These are extra-usage dollars, not quota. How many of these sessions fit inside Pro's or Max's quota is exactly what Devin doesn't publish, and the docs say quota consumption can differ from these rates.
• Model choice swings the cost by about 100 times. The same session costs $0.14 on GPT-6 Luna and $14.25 on GPT-6 Astra, before caching.
• Caching cuts the bill by roughly 70% in this example. Devin's docs say requests to the same model "leverage caching", so switching models mid-task loses that.
• After the free periods end, SWE-2 at its list price would cost the same as Kimi K3 in this example ($4.28 uncached, $1.36 cached), and SWE-2 Flash the same as GPT-6 Luna.
• The same session on Copilot costs the same at list price for the models both list (for example 285 credits, or $2.85, on Claude Sonnet 5.5 uncached). The difference is the packaging: Copilot Pro tells you its allowance is 1,500 credits a month; Devin Pro tells you it has a quota.

To price your own session on Devin Desktop, GitHub Copilot and Cursor side by side, use our AI coding cost calculator. It uses the rates in the table above and shows Devin's part as a range, from $0 if everything fits in the quota to the full extra-usage price if nothing does, since the quota's size isn't published.

Ways to make the quota last

From Devin's quota page and its pricing FAQ:

• Use the free in-house models while they are free. SWE-2 and SWE-2 Flash don't count against the quota until the dates above.
• Use the SWE family or smaller models for routine work. The quota page suggests SWE-1.7 or SWE-1.6, and the pricing FAQ mentions Haiku, GPT 5.2 Mini and Kimi K2.5.
• Keep context small and instructions precise, and avoid long sessions when a quick prompt will do.
• Stay on one frontier model per task so requests hit the cache.
• Try Fusion. It pairs a frontier "lead" model with a cheaper "sidekick" that does the mechanical work, and bills each model at its own rate. The default pairing is Claude Opus 5.5 Medium with SWE-2 Medium. It is on paid plans only.
• Link a ChatGPT plan for GPT models. On Pro, Max and Teams you can link a personal ChatGPT Go, Plus or Pro subscription; requests to GPT models are then billed to ChatGPT and "none of that usage counts against your Devin quota or on-demand credits." Fast and priority GPT variants, non-GPT models and cloud machine time still use Devin quota.

Which plan fits

These follow from the published rules above, not from using the product.

• Mostly Tab completions and inline edits: Free lists both as unlimited.
• Regular agent use on one machine: Pro, plus extra usage when you run out. Keep an eye on the usage meter, because the quota isn't a number you can plan around in advance.
• Heavy days followed by quiet ones: Max has no daily cap, only a weekly one, so a single long day isn't cut off at a daily limit.
• You already pay for ChatGPT Plus or Pro and mostly use GPT models: linking it moves that model usage off your Devin quota.
• A team with a few heavy users and many occasional ones: full seats for the regulars, free flex seats for the rest, and shared on-demand credits for the flex usage.

Might not fit if

• You need to know your allowance in advance. Devin doesn't publish quota sizes. GitHub Copilot lists its monthly allowance in credits; see our Cursor vs GitHub Copilot vs Windsurf comparison.
• You want to try a paid plan first. There are no free trials, only a refund if you haven't started using the subscription.
• Your budget depends on the free models. The SWE-2 and SWE-2 Flash free periods end on November 30 and December 4, 2026.

Check the current numbers

• Devin pricing
• Quota-based usage
• Self-serve plans, seats and on-demand credits
• Models and per-token prices
• Use your ChatGPT plan in Devin

Sources

1. Devin pricing page (windsurf.com/pricing redirects here) — https://devin.ai/pricing, checked November 14, 2026
2. Devin Desktop docs: Quota-based usage — https://docs.devin.ai/desktop/accounts/quota, checked November 14, 2026
3. Devin Desktop docs: Plans and usage — https://docs.devin.ai/desktop/accounts/usage, checked November 14, 2026
4. Devin docs: Self-serve plans — https://docs.devin.ai/admin/billing/self-serve, checked November 14, 2026
5. Devin Desktop docs: AI models (self-serve token rates) — https://docs.devin.ai/desktop/models, checked November 14, 2026
6. Devin Desktop docs: Fusion — https://docs.devin.ai/desktop/fusion, checked November 14, 2026
7. Devin docs: Use your ChatGPT plan in Devin — https://docs.devin.ai/admin/billing/chatgpt, checked November 14, 2026