Lovable vs Bolt.new vs v0: Auth Gaps and Pricing Split in 2026

A crowded market splits into three

The usual way to compare AI app builders is to line up Lovable, Bolt.new and Vercel's v0, check the price tag on each, and pick one. In 2026 that approach misses most of what matters. All three cost about the same to start, but they bill in different units, take very different approaches to authentication, and carry very different security records. Meanwhile, the AI-native code editors that professional developers use, led by Cursor and the product formerly called Windsurf, have turned into tools for managing teams of agents. The app builders now feed into those editors more often than they compete with them.

Most comparisons sort the three builders into the same buckets. Lovable is usually named the best tool for non-technical founders who want a deployed full-stack MVP, because it comes with Supabase authentication and hosting built in31. Bolt.new is the developer's choice for framework flexibility, and v0 is the frontend specialist for teams already on Vercel and Next.js3132. One July analysis argued the three tools are "different species" rather than rival brands, and that most bad reviews come from people picking the wrong kind of tool for their project, not from flaws in the tools themselves39.

Authentication: where the sources disagree

Authentication is where comparisons disagree the most, and the disagreement matters to anyone choosing a tool. Several breakdowns list v0 as having no authentication and no backend313436. Others say Bolt.new and v0 offer it partially32, or that all three now provide auth out of the box37. A comparison published this week says NextAuth or Clerk authentication works with v0 out of the box40. A hands-on April test was more careful: only Lovable produced working auth through Supabase, while v0 needed manual setup and Bolt.new needed manual Clerk or Auth.js wiring38.

The most likely explanation is that the products changed quickly and the reviewers tested different versions. v0 gained Git integration in February 2026, and its output can now arrive as a reviewable pull request39. That pushes it closer to working inside existing codebases where auth is already in place. The practical takeaway still holds, though. Only Lovable reliably builds a complete login system as part of the default flow. For v0 and Bolt.new, plan on wiring it up yourself.

Pricing: similar sticker prices, different meters

Entry prices cluster between $20 and $25 a month, but the reported figures vary. Lovable Pro is generally listed at $25313337, though some guides cite $203540. Bolt.new appears at $20 in some comparisons3234 and $25 for 10 million tokens in others313337. v0 Premium is consistently $203137.

The headline price matters less than what each plan actually meters. Lovable sells credits per task: a small styling tweak costs about half a credit, and a landing page with images about 1.739. Bolt.new sells raw tokens. Several analyses warn that token use per message grows with project size, so the same edit gets more expensive as an app grows39. Users have reported monthly bills approaching $500 to $1,000 once projects grew past the prototype stage33. v0's metered model ties cost to model choice and output size, so you often don't know what a generation costs until it runs39. v0 also faced backlash over metered pricing in May 202534.

Free tiers follow the same pattern. Bolt.new is generally considered the most generous at around 1 million tokens a month3137. Lovable offers about five credits a day with a monthly cap of roughly 303337. Treat any single price as approximate, and pay closer attention to how each tool's costs grow with heavy use.

Security: Lovable's pattern of incidents

The biggest risk in this category is that apps can be easy to build and easy to breach at the same time. Lovable is the most visible example, partly because it is the largest player. It is reportedly valued at $6.6 billion, with about 8 million users and roughly $400 million in annual recurring revenue as of February 202633.

The problems go back to 2025. Researchers at rival Replit scanned 1,645 apps showcased on Lovable and found 170 that let anyone access user data16. The cause was Supabase databases connected without row-level security policies12. The finding became CVE-2025-48757, which Lovable disputes, blaming users instead12. Lovable's response at the time was a pre-publish security scan. According to the vulnerability report, the scan only checked whether access controls were turned on, not whether they were configured correctly16.

In 2026 the incidents came from Lovable's own platform, not just from apps built on it. In February, a researcher found 16 vulnerabilities, six of them critical, in one app featured on Lovable's Discover page. Its authentication logic was inverted, blocking logged-in users while letting anonymous visitors through, and 18,697 user records were exposed1119. The researcher's support ticket was initially closed, though Lovable's security team later got in touch19.

In April came a broken object-level authorization flaw in Lovable's API. Any free account could reach another user's source code, database credentials and AI chat histories1113. It was reported through HackerOne on March 3. Lovable patched it only for projects created after November 2025, so older projects stayed exposed for about 48 days, and a follow-up report was closed as a duplicate13. Lovable's public response moved from denying a breach, to blaming unclear documentation, to blaming HackerOne, to a partial apology11. The company admitted that during a February backend change it had accidentally re-enabled access to chats on public projects15.

The framing of the incident is disputed. ESET's Jake Moore described it as closer to a design flaw than a traditional breach, while warning that vibe coding keeps speeding up bad defaults15. Computing noted that because Lovable apps often embed credentials for Supabase, Stripe and Google APIs, the exposure could spread to other services13. That second point is the stronger one. Users who managed their own secrets carefully could still be exposed by a flaw in the platform itself. No amount of personal care fixes that.

The broader numbers are sobering, though they come mostly from secondary sources. Reporting cites studies finding vulnerabilities in 40% to 62% of AI-generated code. One first-quarter 2026 assessment found that 91.5% of more than 200 vibe-coded apps had at least one flaw traceable to AI hallucination11. An academic paper from June 2026 also documented a deployed healthcare app that exposed customer data through broken access control18. Lovable has since partnered with Aikido on automated penetration testing11. Even so, the record shows that faster patching hasn't changed the defaults underneath.

Cursor becomes a place to run agents

The guidance across these comparisons is consistent. Builders are good for validating an idea, and serious products eventually move into tools like Cursor or Claude Code39. That makes the professional editors part of the same story.

Cursor's direction is now clear. Version 2.0 introduced Composer, its own coding model, along with an interface built around agents rather than files21. Version 2.4 in January 2026 added subagents that run in parallel, each with its own context window, plus Skills, which are reusable task instructions2229. Cursor 3.0 arrived on April 2, 2026 with an Agents Window for running work locally, in the cloud or over SSH. Version 3.1 added tiled panes for running several agents at once30. Composer 2 shipped in March and Composer 2.5 in May29. Cloud agents can be started from Slack, GitHub, Linear or the API30.

Cursor's growth figures vary from source to source. One reports 10 million active developers in January22, while another cites more than $500 million in annual recurring revenue5. A September comparison says Cursor is reportedly being acquired for about $60 billion2. No other coverage here confirms that, so it should be treated as unverified.

Windsurf is gone, and Devin Desktop replaces it

The product once called Windsurf has changed the most. Accounts of the ownership history conflict. Some date Cognition's acquisition to July 2025, right after OpenAI's deal collapsed and Google paid $2.4 billion to license the technology and hire the founders89. Others say December 202513 or late March 20267. The July 14, 2025 agreement is the date most often cited89.

The later events are less disputed. Cognition replaced credits with daily and weekly quotas on March 19, 2026, raising Pro from $15 to $20, the same as Cursor1. Existing subscribers kept their legacy pricing2. Windsurf 2.0 arrived on April 154. On June 2, an over-the-air update renamed the editor Devin Desktop23. The Cascade agent was retired on July 1 and replaced by Devin Local, which Cognition says is up to 30% more token-efficient9. The product now includes an Agent Command Center that tracks local and cloud agents on a Kanban board, the SWE-1.6 model, and support for the open Agent Client Protocol3.

The verdict

Both parts of the market are heading the same way: the AI does more of the work, and humans review the results. For professional editors, that has produced better review tools, sandboxing and enterprise controls. For app builders aimed at non-coders, the people making the software are often the least equipped to review it. Lovable's growth and Lovable's security record come from the same source: it makes deploying with real data effortless. Until secure defaults are enforced rather than optional, the sensible approach is to use these builders to test ideas, rotate any credential that has passed through a builder chat, and move anything with real users into a proper editor with a security review.