Skyvern pricing, decoded: what a credit actually buys, and where it stops.

Free at $0/month with 1,000 credits, Hobby at $29/month with 30,000 credits, Pro at $149/month with 150,000 credits, and Enterprise at custom pricing with unlimited credits and self-hosted deployment. The structure is published on skyvern.com/pricing. The Free tier includes CAPTCHA solving and basic support, Hobby adds priority support, faster execution, and webhook integrations, Pro adds team workspaces, advanced workflows, and 2FA credential management, and Enterprise adds HIPAA, SOC 2 Type II, SSO, a dedicated account manager, and SLA guarantees.

Skyvern does not publish a per-action rate, but the public tier ratios let you back one out. The Day 5 launch post on 30 January 2026 says Hobby gives roughly 1,200 actions for 30,000 credits, which is about 25 credits per action. At Hobby that is roughly $0.024 per action. At Pro (150,000 credits, roughly 6,200 actions for $149), the implied cost lands in the same neighborhood, around $0.024 per action. The official caveat is that a single run consumes credits based on complexity and duration, so any per-action number is a tier-average rather than a guaranteed unit price.

Skyvern's previous public model was a flat $0.05 per step. The Day 5 launch post on 30 January 2026 announced the move to monthly credits, with the stated reason that per-step billing forced users to think in the wrong units and discouraged workflow improvements that required adding steps. Credits were positioned as a unit of browser execution that absorbs runtime, page complexity, retries, and anti-bot work into one number, instead of charging separately per click. From the buyer's side, the new model trades a transparent per-step rate for a tier-bundled rate that is harder to predict per workflow but cheaper if your usage is steady.

The Day 5 launch post says credits represent a unit of browser execution, and that different workflows consume different amounts depending on runtime, page complexity, retries, and anti-bot measures (CAPTCHA, proxies, geo-targeting). One credit is not a step, not a page, and not a second. It is a synthetic unit that bundles cloud compute, the vision LLM call, the proxy or anti-bot allowance, and any retries the runtime needed. That is why two workflows with the same number of clicks can burn different credit amounts.

No. The Skyvern repository at github.com/Skyvern-AI/skyvern, licensed AGPL-3.0, is browser-only. The runtime is built on a Playwright-compatible SDK and a vision LLM that reads the rendered Chromium viewport. There is no Windows desktop runtime, no Citrix runtime, and no mainframe terminal connector. If your workflow has to drive a SAP GUI window, an Oracle Forms session, a Jack Henry green-screen, or an Epic Hyperspace patient chart, the credit unit cannot price it because the surface the credit measures is not where your work happens.

The open-source repo is AGPL-3.0 and you can run it on your own infrastructure. The cloud product's anti-bot measures stay proprietary, so the self-hosted version covers the core runtime but not the production-grade CAPTCHA and proxy tooling. Self-hosting also means you carry the cost of the vision LLM yourself and you size your own concurrency, instead of buying it bundled in a tier. Self-hosting is a credible answer for steady, predictable workloads where the bundled cloud features are not load-bearing. It is not the answer if you need the managed CAPTCHA and proxy fabric that the credit price subsidizes.

The published action equivalents from the launch post are roughly 170 actions per month on Free, roughly 1,200 on Hobby, and roughly 6,200 on Pro. These are not contractual limits, they are working approximations the Skyvern team posted as guidance. A single action that triggers a CAPTCHA, a retry, or a heavy anti-bot path will burn more credits than the average. A simple click on a static page burns less. The action numbers are useful as a starting estimate for sizing a tier, not as a billing guarantee.

Skyvern's own marketing draws the comparison: traditional RPA charges roughly $10,000 per bot per year with 20 to 40 percent monthly maintenance, while Skyvern uses usage-based credits with low maintenance overhead. The framing is fair for browser-tab workflows, where the model is genuinely cheaper and the maintenance burden is lower because vision-based selection is more resilient to layout drift. The framing breaks down for desktop workflows that traditional RPA does cover and Skyvern does not, because there is no surface for the credit to measure on a SAP GUI window.

A credit is a unit of managed-cloud browser execution: it bundles the VM hour, the vision call, the proxy, and the retry budget into one synthetic number. A desktop RPA agent runs on the user's own Windows session, drives the operating system through the UI Automation accessibility tree, and bills against wall-clock minutes the agent is active. There is no managed VM to amortize, no vision LLM round-trip per click, and no proxy fabric. The natural unit is time-on-desktop. Mediar prices that unit explicitly at $0.75 per minute, drawn against a $10,000 program prepay that converts to credits.

It makes sense if your workflow lives entirely inside a Chromium tab, your usage is steady enough to fit a tier, and you value the bundled CAPTCHA and proxy work that the credit price subsidizes. Vendor portal logins, lead enrichment, payer claim status checks, document downloads, web SaaS form fills. It does not make sense if your workflow crosses out of the tab into a desktop window, or if your usage is spiky enough that the tier ceiling forces you onto Enterprise unnecessarily. Match the unit you are billed in (credits, minutes, per-bot license) to the surface your workflow actually lives on, and the price will tell you the truth about whether the tool fits.