Lovable vs Bolt vs v0: I tested the same app on all three (and the credit price is the smallest problem)

The comparison that matters never shows three landing pages side by side. It shows one barbershop app, the same 11 criteria, built in three different builders, with bills and delivery on record. In that test documented by zilvodev on TabNews on Feb 26, 2026, Bolt closed with the lowest total bill and the worst proportional delivery, while the highest cost per attempt turned into the lowest cost per met criterion. The credit price alone told the story backwards.

This post organizes what three independent sources show on equal challenges: the barbershop test with 11 criteria (TabNews, Feb 26, 2026), the 8-month tracking across 6 builders (socialanimal), the per-tool cuts (ibe, Jun 4, 2026), and the 12-project production analysis (Future Humanism, Feb 17, 2026). For the cluster's full math, with credits, tokens, and monthly operation, return to the what vibe coding really costs guide.

The barbershop test flips hasty price reads in a few lines. Bolt cost R$3.21 total and met 4 of 11 criteria, or R$0.80 per met criterion. Lovable cost R$4.25 and met 8 of 11, or R$0.53 per criterion. Emergent cost R$4.59 and met 11 of 11, or R$0.42 per criterion (TabNews, Feb 26, 2026). Every number comes from one report, one app, one 11-criterion ruler, so delivery per real compares with no mixed scopes.

The hands-on consequence lands when picking the best vibe-coding tool for a real case. Whoever watches only the total picks Bolt and receives under half the criteria. Whoever divides by what shipped finds Emergent most efficient per unit and Lovable in the middle, with double Bolt's yield for about one real more overall. Cheap per attempt never means cheap per result, and this test measures the gap exactly.

Table with barbershop-test numbers (TabNews, Feb 26, 2026), per-tool cuts (ibe, Jun 4, 2026), and 8 months across 6 builders (socialanimal).

The barbershop test followed the genre's most honest rule: one app, one 11-criterion set, different builders, cost and delivery logged per round (TabNews, Feb 26, 2026). The report also measures the loop, with 211k tokens burned across 3 fixes with nothing repaired, against 39k tokens at initial build, a 5.4x multiple turned into the doom-loop portrait. And it logs one behavioral detail outweighing any price table: none of the tools checked their own work before shipping.

The methodology caveat comes from the report itself and must travel with any ranking. The builders ran distinct models, with Sonnet 4.5 on one side and 4.6 named on the other, which blocks treating the gap as pure product superiority (TabNews, Feb 26, 2026). Different models, with distinct windows, costs, and behaviors, contaminate direct comparison. The ranking holds as a strong efficiency-per-delivery signal in that cut, never a final sentence for every project.

What holds the read beyond one isolated test

Two sources widen the base past the barbershop. The socialanimal survey tracked 6 builders for 8 months, with plan tables, $200-500 monthly production, and $50-200K legacy Bubble exits (socialanimal). The Future Humanism analysis watched 12 projects with auth breaking in production, debug near $15k against initial builds near $6k (Future Humanism, Feb 17, 2026). One test measures delivery per real short-term. The other two measure what happens once the app must live publicly with real users.

The heart of the same-app Lovable-versus-Bolt test is yield, never total spend. Lovable at 8 of 11 shows a nearly whole app, with pointed gaps. Bolt at 4 of 11 shows a start that stalled midway, over half the criteria pending. Emergent at 11 of 11 shows full delivery on a closed scope (TabNews, Feb 26, 2026). All three totals sit in one order of magnitude, R$3 to R$5, and deliveries sit on distinct tiers. The question stops being what attempts cost and turns into what each working criterion costs.

The chart below translates that inversion into one image. Each bar is cost per met criterion, computed from the report's numbers. The shortest bar marks the most efficient delivery per real, never the cheapest tool overall.

A cold read of all three outcomes suggests one plain budgeting rule. Closed scopes with few pending criteria take the middle builder, provided someone reviews before publishing. Scopes needing whole delivery in cycle one want the top-yield builder, even at a slightly higher total. And demos that only must show flow on screen can live with low yield, provided nobody mistakes the demo for the finished app. The common mistake treats 4 of 11 as a discount, when it really is remaining work billing more tokens ahead.

The Bolt-versus-v0 comparison only makes sense with the chore named, because both were born for distinct regimes. The cut documented by ibe on Jun 4, 2026 compresses to one line each: Lovable for full apps, Bolt for fast demos, v0 for one-off components (ibe, Jun 4, 2026). Each builder earns most on the chore matching its shape. Forcing a demo into a full system down one thread is the shortest road to the fix loop.

Architecture explains part of the behavior. Bolt runs generation in the browser through WebContainers, favoring visible, fast iteration, with instant preview and short tweak cycles (ibe, Jun 4, 2026). v0 assumes an existing project, generates the one-off component, and hands the snippet back to paste into the stack, which works once routing, styling, and data already resolve at the destination. Lovable aims at the full app, screens, flow, and integrations in one place, which explains the higher barbershop yield plus the wider surface for auth and backend to fail later.

the credits-and-tokens math

In practice, the per-chore pick runs like this: validating an offer end to end on plain data wants the full-app builder. Aligning client expectations, showing a clickable flow, or testing one screen in a meeting wants the demo builder. Unblocking one interface inside an existing project wants the component generator. When chores and tools cross wrong, the symptom shows fast as rework, bloated context, and criteria that never close.

The reports' most uncomfortable number is no list price. It is 211k tokens burned across 3 fixes with nothing repaired, against 39k tokens at initial build, a 5.4x multiple (TabNews, Feb 26, 2026). That is the doom loop in portrait: the app looking nearly done, every tweak promising to close, every fresh attempt returning the defect in new wording. The initial build cost a fraction. Stubbornness down one thread cost the multiple.

The behavior feeding the loop sits in the same report, blunt: none of the tools checked their own work before shipping (TabNews, Feb 26, 2026). With no verification stage, mistakes walk straight to the user, who pays another round to point at the obvious. The 8-month tracking shows the same trouble's other faces at bigger scale: 847 dead lines found in a Bolt-generated project plus one case where Cursor rewrote working code into broken code (socialanimal). Dead code weighs on context and maintenance. Rewrites that break turn working code into fresh bugs to fix with more tokens.

total-cost comparisons hide that 4 of 11 leaves 7 criteria for the loop. When every fix reloads context and no tool verifies its own work, the remainder costs more than the start, which explains the 5.4x multiple between fixing and building.

The defense against that pattern is procedural, never buying more credits. Freezing scope once yield stalls, isolating the troubled module in a fresh thread, and reviewing outside the generator before publishing break the documented cycle. Insisting down one thread, with one poisoned context, swaps a small build bill for a big repetition bill.

Finished screens never mean finished apps, and production reports converge there with hard numbers. The 8-month tracking found auth middleware plus Stripe in Lovable running 3 distinct patterns, a sign auth and billing got improvised instead of contracted (socialanimal). Three patterns for one responsibility mean three places to audit, three surfaces to break, three flows to maintain. In demos that passes unseen. With real users and money involved, it becomes incidents.

The 12-project analysis prices the risk: auth breaking in production, debug near $15k against initial builds near $6k (Future Humanism, Feb 17, 2026). Building cost a fraction. Fixing with locked-out users cost the multiple. Monthly operation completes the frame, typical production at $200-500 a month plus $50-200K legacy Bubble exits, besides the plan table the survey details (socialanimal). Validating costs little. Operating with login, permissions, and billing demands budget plus a review rite.

where vibe coding breaks on auth and backend

For anyone eyeing Lovable code exports to maintain outside, the reports' message runs sober and useful. The cited sources measure no single export number, so any frictionless-migration promise invents. What the reports do measure are the frictions around leaving: duplicated middleware to unify, dead lines to prune, plan tables to compare, legacy costs to weigh. Exporting fits these tools' shape, and the exit cost follows auth and backend state at decision time, never the download button.

The call simplifies once it starts from the chore plus the code's destination, never the tool logo.

When Lovable makes most sense

Lean full apps validating an offer end to end, with screens, flow, and plain data. The 8-of-11 barbershop yield backs that cut (TabNews, Feb 26, 2026). The boundary condition comes from the 8-month tracking: treat auth, Stripe, and permissions as reviewed spec, never as improv inside the screen prompt (socialanimal).

When Bolt makes most sense

Fast demos aligning clients, testing flows in meetings, or iterating visuals on short cycles, riding browser WebContainers (ibe, Jun 4, 2026). The 4-of-11 barbershop score marks the limit: good for showing, never for operating (TabNews, Feb 26, 2026). If the demo takes off, next is freezing, reviewing the 847 dead lines from the 8-month tracking, and splitting the core before publishing.

When v0 makes most sense

One-off components inside existing projects, with routing, styling, and data settled at the destination (ibe, Jun 4, 2026). Since v0 assumes a project, the Bolt-versus-v0 comparison settles on the starting-point question: with no project, the component generator has nowhere to live. With a project, it unblocks screens without dragging the whole app into the prompt.

Which runs cheapest: Lovable, Bolt, or v0?

In the 11-criterion barbershop test, the lowest total was Bolt at R$3.21, then Lovable at R$4.25, Emergent at R$4.59 (TabNews, Feb 26, 2026). Per met criterion the order flips: Emergent at R$0.42 (11/11), Lovable at R$0.53 (8/11), Bolt at R$0.80 (4/11). v0 never entered that test, so no comparable number exists in that cut. The honest answer splits both rulers: smallest upfront spend and lowest cost per delivery are distinct titles.

What is the best vibe-coding tool for a full app?

For lean full apps, the reports point at Lovable for the cut, with 8 of 11 on one challenge and a shape aimed at screens, flow, and integrations in one place (TabNews, Feb 26, 2026; ibe, Jun 4, 2026). The caveat travels with the recommendation: auth middleware plus Stripe in 3 patterns across the 8-month tracking shows login and billing want spec plus review (socialanimal). Full apps with no auth review are demos with expiry dates.

Can I export Lovable code and maintain it outside the builder?

The cited sources carry no single export measure with a number, so distrust round promises on the topic. What the reports document are the costs around leaving: typical production at $200-500 a month, $50-200K legacy Bubble exits, plan tables for comparing regimes (socialanimal). Add the code state at exit time, with duplicated middleware and dead lines as examples of what wants cleanup before migrating. Exporting belongs in the plan from day one, with auth and backend documented outside the prompt.

Why do auth and backend break so often in vibe coding?

Because they tolerate improvising least and suffer bloated context most. The barbershop test shows the mechanism with 211k tokens across 3 unrepaired fixes against 39k at start, or 5.4x, with zero verification before shipping (TabNews, Feb 26, 2026). Production shows the price with 12 projects and auth blowing live, debug near $15k against builds near $6k (Future Humanism, Feb 17, 2026). The repeated exit in the reports is speccing login, sessions, permissions, and billing as reviewed spec, with deploys locked until verification.

The same-app test leaves three lessons one line each. Total cost with no yield deceives, because R$3.21 at 4 of 11 runs R$0.80 per delivery while R$4.59 at 11 of 11 runs R$0.42 (TabNews, Feb 26, 2026). Cuts decide more than brands, Lovable for full apps, Bolt for fast demos, v0 for one-off components (ibe, Jun 4, 2026). And production bills in auth, with tripled middleware, 847 dead lines, 12 projects with login blowing live, $15k debug versus $6k build (socialanimal; Future Humanism, Feb 17, 2026).

The next step follows the remaining doubt. For the credit-and-token math behind the loop, advance to the detailed consumption math. For where login and backend die live, advance to the auth and backend guide. For the month's full math, with operation and rework, return to the cluster pillar.

• TabNews, zilvodev, Feb 26, 2026: barbershop app with 11 criteria, Lovable R$4.25 at 8/11 (R$0.53 per criterion), Bolt R$3.21 at 4/11 (R$0.80), Emergent R$4.59 at 11/11 (R$0.42), 211k tokens across 3 unrepaired fixes vs 39k at start (5.4x), doom loop, no verification before shipping, distinct-models caveat Sonnet 4.5 vs 4.6.
• socialanimal: 8 months across 6 builders, Lovable auth and Stripe middleware in 3 patterns, Bolt with 847 dead lines, Cursor rewriting working code into broken code, plan tables, production at $200-500 a month, $50-200K legacy Bubble exits.
• ibe, Jun 4, 2026: cuts with Lovable for full apps, Bolt for fast demos, v0 for one-off components, Bolt on WebContainers, v0 assuming an existing project.
• Future Humanism, Feb 17, 2026: 12-project analysis with auth breaking in production, debug near $15k vs initial builds near $6k.