# AI app builder benchmarks: evidence and evaluation protocol

Canonical: https://uvilno.com/benchmarks
Updated: 2026-10-28

Published by Uvilno. Uvilno is presented first as the publisher's editorial order, not an independent rank. Documentation, recorded first-party checks and unmeasured outcomes are distinct. Overall quality scores remain unmeasured. Prices and vendor capabilities can change; source access and operating services are separate purchases.

## 2026 full-stack engineering evaluation rubric

Uvilno first because Uvilno publishes this comparison; not a measured rank. The five publisher-selected criteria total 100 points. A documented capability is not an equivalent hands-on result. Unknown checks stay unmeasured, never failed. Overall scores remain null until a comparable test protocol is completed.

- **Code ownership & export cost: 25 possible points.** Build the same app, obtain its complete source, record the actual purchase, and rebuild on a separate host. Test continued source access and document the backend services that still need migration.
- **Database, accounts & app AI: 25 possible points.** Create, edit and remove the same records; reload; repeat with another account; check record permissions; run a real app-AI request and inspect a triggered automation. Record each managed or external dependency.
- **Targeted edit precision & speed: 20 possible points.** Apply the same specified changes to matched projects. Compare exact file deltas, verify the rendered target and unchanged routes, and record end-to-end timings including verification. Publish failures as well as successes.
- **No-usage editing & recovery: 15 possible points.** Make the same direct edit and introduce the same reproducible fault. Inspect actual usage ledgers, repair limits and saved-version behavior. A limited free path must not be described as unlimited free coding.
- **Multi-page & interactive applications: 15 possible points.** Build and navigate the same route plan, exercise state across pages, then measure frame time and input response on named devices. Framework support or an attractive screenshot does not establish a universal 60 FPS result.

### Uvilno

Overall score: not measured. 17 documented or recorded checklist conditions; 3 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: An owned, unlocked project can export its runtime-appropriate source and referenced media. Sources: https://uvilno.com/docs.
- Documented: Supported frontends can use authorized GitHub import, push and pull; a source push requires unlock. Sources: https://uvilno.com/docs.
- Documented: Free is $0; a source unlock is $4.99 once. Plus is $10/month, Pro $20/month and Ultra $59/month. Generation and hosted services are separate. Sources: https://uvilno.com/pricing.
- Documented: The $4.99 purchase permanently unlocks one project's source; subscription generation remains finite. Sources: https://uvilno.com/pricing.
- Documented: Source export is not a copy of managed records, visitor identities, payment credentials or hosted services. Sources: https://uvilno.com/docs.

**Database, accounts & app AI**
- Documented: Uvilno.db CRUD and App Data use schema-declared, persistent app records. Sources: https://uvilno.com/benchmarks.
- Documented: Managed visitor identities are isolated per project from the creator's platform session. Sources: https://uvilno.com/benchmarks.
- Documented: Uvilno.ai.ask/json makes real owner-funded calls; admission limits and a 600-output-token ceiling apply. Sources: https://uvilno.com/benchmarks.
- Documented: data/schema.json declares collection fields and access; the runtime exposes validated data operations. Sources: https://uvilno.com/docs.
- Documented: Bounded record effects and notifications execute in the triggering record transaction; arbitrary server code, email and network actions are outside this rules engine. Sources: https://uvilno.com/benchmarks.

**Targeted edit precision & speed**
- Documented: @file, @page, @db and bounded repository retrieval supply explicit context. Sources: https://uvilno.com/docs.
- Documented: The mapped inspector supplies selected-element context or a supported direct text, link or color change. Sources: https://uvilno.com/docs.
- Unmeasured: Recorded targeted checks do not establish a comparative edit-accuracy rate across vendors.
- Unmeasured: No matched cross-vendor timing establishes a universal 4–8 second edit duration.

**No-usage editing & recovery**
- Documented: Mapped direct text, link and color changes use 0% model usage; unsupported computed elements use the normal scoped-edit path. Sources: https://uvilno.com/docs.
- Documented: A reproduced native-static runtime fault was repaired and rechecked without customer usage. This bounded path is rate-limited and does not cover every engine error. Sources: https://uvilno.com/benchmarks.
- Documented: Saved-version and entitled per-file restores preserve checkpoints without invoking the generation model; verification still applies. Sources: https://uvilno.com/docs.

**Multi-page & interactive applications**
- Documented: Contract-first route plans and reusable modules support multi-page frontends; saved legacy and eligible Next.js projects retain their respective runtimes. Sources: https://uvilno.com/docs.
- Documented: The portable Canvas game runtime and authored interactive fixtures provide workload evidence, not a promise about every generated game. Sources: https://uvilno.com/benchmarks.
- Unmeasured: A matched hardware/workload frame-time study has not been published; 60 FPS is not guaranteed.

### Emergent

Overall score: not measured. 7 documented or recorded checklist conditions; 13 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: Save source through GitHub; copy individual files in Code view. Sources: https://help.emergent.sh/your-data-ownership.
- Documented: GitHub source handoff on Standard and above. Sources: https://help.emergent.sh/your-data-ownership.
- Unmeasured: The displayed Standard offer is $20/month equivalent under Annual billing ($240/year). A month-to-month checkout price is not verified here. An equivalent complete source-access purchase has not been priced by a matched export test.
- Unmeasured: This specific source-access condition has not been established here; it is not marked unsupported.
- Documented: Exported code excludes secrets and application data; migrate those separately. Sources: https://help.emergent.sh/your-data-ownership.

**Database, accounts & app AI**
- Documented: Backend and database workflows are part of the app platform. Sources: https://help.emergent.sh/faqs.
- Documented: Generated app login requires configuration and testing. Sources: https://help.emergent.sh/faqs.
- Documented: One-click LLM integration is listed; app configuration and usage still apply. Sources: https://emergent.sh/pricing.
- Unmeasured: An equivalent schema-backed SDK contract has not been inspected here.
- Unmeasured: This automation condition has not been established here; supported integrations may provide it.

**Targeted edit precision & speed**
- Unmeasured: Exact context-selection behavior needs a documented or observed matched test; repository access alone does not prove surgical precision.
- Unmeasured: Equivalent selected-element context has not been checked here.
- Unmeasured: No equivalent before/after edit-accuracy benchmark is recorded here.
- Unmeasured: No comparable end-to-end edit-timing distribution is recorded here.

**No-usage editing & recovery**
- Unmeasured: No-usage visual-edit terms have not been established here; absence of a recorded check is not a paid-only claim.
- Unmeasured: The same injected fault and actual customer usage ledger have not been inspected on this product.
- Unmeasured: A matched restore and usage-ledger check has not been run here; existing source-control features are described in the product comparison.

**Multi-page & interactive applications**
- Documented: Agent-led full-stack web and mobile app building. Sources: https://help.emergent.sh/faqs.
- Unmeasured: An equivalent interactive workload has not been run in this evaluation; this does not imply an inability to build games.
- Unmeasured: No matched workload/device frame-time measurement is recorded here.

### Base44

Overall score: not measured. 9 documented or recorded checklist conditions; 11 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: Advertises ownership and two-way GitHub source export; plan conditions apply. Sources: https://base44.com/pricing.
- Documented: GitHub integration on Builder and above; Base Code edits connected repositories. Sources: https://docs.base44.com/Getting-Started/base-code.
- Unmeasured: Starter is displayed at $16/month billed annually ($192 upfront). Builder, which lists GitHub integration, is $40/month billed annually ($480 upfront). An equivalent complete source-access purchase has not been priced by a matched export test.
- Unmeasured: This specific source-access condition has not been established here; it is not marked unsupported.
- Documented: Source ownership is documented; migrate backend data and connected services separately. Sources: https://docs.base44.com/Getting-Started/base-code.

**Database, accounts & app AI**
- Documented: Built-in entities, records and a data dashboard; test and production data controls. Sources: https://docs.base44.com/Building-your-app/Managing-your-app-data.
- Documented: App sign-in and role/access settings; data permissions require review. Sources: https://docs.base44.com/Setting-up-your-app/Managing-access.
- Documented: invokeLLM and app agents use supported managed model presets. Sources: https://docs.base44.com/developers/backend/resources/models.
- Documented: Typed entities and visual record management are part of the built-in backend. Sources: https://docs.base44.com/Building-your-app/Managing-your-app-data.
- Documented: Scheduled, data-event and connector triggers; workflows are replacing automations. Sources: https://docs.base44.com/Building-your-app/Creating-automations.

**Targeted edit precision & speed**
- Unmeasured: Exact context-selection behavior needs a documented or observed matched test; repository access alone does not prove surgical precision.
- Unmeasured: Equivalent selected-element context has not been checked here.
- Unmeasured: No equivalent before/after edit-accuracy benchmark is recorded here.
- Unmeasured: No comparable end-to-end edit-timing distribution is recorded here.

**No-usage editing & recovery**
- Unmeasured: No-usage visual-edit terms have not been established here; absence of a recorded check is not a paid-only claim.
- Unmeasured: The same injected fault and actual customer usage ledger have not been inspected on this product.
- Unmeasured: A matched restore and usage-ledger check has not been run here; existing source-control features are described in the product comparison.

**Multi-page & interactive applications**
- Documented: Integrated AI app building, hosted backend and data management. Sources: https://base44.com/pricing.
- Unmeasured: An equivalent interactive workload has not been run in this evaluation; this does not imply an inability to build games.
- Unmeasured: No matched workload/device frame-time measurement is recorded here.

### Lovable

Overall score: not measured. 9 documented or recorded checklist conditions; 11 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: Export source and move hosting; services require separate migration. Sources: https://docs.lovable.dev/tips-tricks/deployment-hosting-ownership.
- Documented: Two-way GitHub, GitLab and Bitbucket sync; existing-repo import differs. Sources: https://docs.lovable.dev/integrations/git-sync-overview.
- Unmeasured: Pro 100: $25 billed monthly, or $250 paid for a year. Annual payment is upfront. An equivalent complete source-access purchase has not been priced by a matched export test.
- Unmeasured: This specific source-access condition has not been established here; it is not marked unsupported.
- Documented: Code and data can move; replace or reconfigure backend, AI and connector calls. Sources: https://docs.lovable.dev/tips-tricks/deployment-hosting-ownership.

**Database, accounts & app AI**
- Documented: Built-in Cloud backend with database, storage and compute billing. Sources: https://docs.lovable.dev/introduction/credits-and-usage.
- Documented: App authentication is part of the backend configuration. Sources: https://docs.lovable.dev/tips-tricks/deployment-hosting-ownership.
- Documented: Managed app AI connector provides real model calls and workspace usage billing. Sources: https://docs.lovable.dev/features/ai.
- Unmeasured: An equivalent schema-backed SDK contract has not been inspected here.
- Unmeasured: This automation condition has not been established here; supported integrations may provide it.

**Targeted edit precision & speed**
- Unmeasured: Exact context-selection behavior needs a documented or observed matched test; repository access alone does not prove surgical precision.
- Documented: The preview toolbar attaches one or more selected elements to the chat as edit context; sending a requested change consumes standard build usage. Sources: https://docs.lovable.dev/features/preview-toolbar.
- Unmeasured: No equivalent before/after edit-accuracy benchmark is recorded here.
- Unmeasured: No comparable end-to-end edit-timing distribution is recorded here.

**No-usage editing & recovery**
- Documented: The preview toolbar includes 100 free inline text edits per account, replenished after 24 hours, with a shared 2,000-edit workspace limit per 24 hours. Further edits use standard credits. Sources: https://docs.lovable.dev/features/preview-toolbar.
- Unmeasured: The same injected fault and actual customer usage ledger have not been inspected on this product.
- Unmeasured: A matched restore and usage-ledger check has not been run here; existing source-control features are described in the product comparison.

**Multi-page & interactive applications**
- Documented: AI builder for websites and full-stack web apps. Sources: https://docs.lovable.dev/tips-tricks/deployment-hosting-ownership.
- Unmeasured: An equivalent interactive workload has not been run in this evaluation; this does not imply an inability to build games.
- Unmeasured: No matched workload/device frame-time measurement is recorded here.

### Bolt.new

Overall score: not measured. 6 documented or recorded checklist conditions; 14 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: Download a project or use GitHub for source handoff. Sources: https://support.bolt.new/integrations/git.
- Documented: GitHub import, sync and branch workflows. Sources: https://support.bolt.new/integrations/git.
- Unmeasured: Entry Pro is $25 billed monthly. No unverified annual discount is applied. An equivalent complete source-access purchase has not been priced by a matched export test.
- Unmeasured: This specific source-access condition has not been established here; it is not marked unsupported.
- Documented: Source can be self-hosted; migrate database and service dependencies too. Sources: https://support.bolt.new/integrations/git.

**Database, accounts & app AI**
- Documented: Bolt Cloud databases and supported external providers. Sources: https://bolt.new/pricing.
- Documented: Authentication integrations are part of the app-building workflow. Sources: https://support.bolt.new/building.
- Unmeasured: App-user inference has not been established in this source set; builder AI alone is not treated as app AI.
- Unmeasured: An equivalent schema-backed SDK contract has not been inspected here.
- Unmeasured: This automation condition has not been established here; supported integrations may provide it.

**Targeted edit precision & speed**
- Unmeasured: Exact context-selection behavior needs a documented or observed matched test; repository access alone does not prove surgical precision.
- Unmeasured: Equivalent selected-element context has not been checked here.
- Unmeasured: No equivalent before/after edit-accuracy benchmark is recorded here.
- Unmeasured: No comparable end-to-end edit-timing distribution is recorded here.

**No-usage editing & recovery**
- Unmeasured: No-usage visual-edit terms have not been established here; absence of a recorded check is not a paid-only claim.
- Unmeasured: The same injected fault and actual customer usage ledger have not been inspected on this product.
- Unmeasured: A matched restore and usage-ledger check has not been run here; existing source-control features are described in the product comparison.

**Multi-page & interactive applications**
- Documented: Browser-based AI website and application development. Sources: https://support.bolt.new/building.
- Unmeasured: An equivalent interactive workload has not been run in this evaluation; this does not imply an inability to build games.
- Unmeasured: No matched workload/device frame-time measurement is recorded here.

### Replit Agent

Overall score: not measured. 6 documented or recorded checklist conditions; 14 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: Plan source, dependencies and service data as separate handoff items. Sources: https://docs.replit.com/billing/about-usage-based-billing.
- Documented: Development workspace for maintaining application code. Sources: https://docs.replit.com/billing/plans/replit-core.
- Unmeasured: The current official pricing page did not expose a verifiable Core amount in accessible text. Enter your confirmed current monthly quote below. An equivalent complete source-access purchase has not been priced by a matched export test.
- Unmeasured: This specific source-access condition has not been established here; it is not marked unsupported.
- Documented: A migration must include runtime, environment variables and production data. Sources: https://docs.replit.com/billing/about-usage-based-billing.

**Database, accounts & app AI**
- Documented: Production databases, compute and network resources are metered. Sources: https://docs.replit.com/billing/about-usage-based-billing.
- Documented: App authentication requires a configured and reviewed implementation. Sources: https://docs.replit.com/billing/plans/replit-core.
- Unmeasured: App-user inference has not been established in this source set; builder AI alone is not treated as app AI.
- Unmeasured: An equivalent schema-backed SDK contract has not been inspected here.
- Unmeasured: This automation condition has not been established here; supported integrations may provide it.

**Targeted edit precision & speed**
- Unmeasured: Exact context-selection behavior needs a documented or observed matched test; repository access alone does not prove surgical precision.
- Unmeasured: Equivalent selected-element context has not been checked here.
- Unmeasured: No equivalent before/after edit-accuracy benchmark is recorded here.
- Unmeasured: No comparable end-to-end edit-timing distribution is recorded here.

**No-usage editing & recovery**
- Unmeasured: No-usage visual-edit terms have not been established here; absence of a recorded check is not a paid-only claim.
- Unmeasured: The same injected fault and actual customer usage ledger have not been inspected on this product.
- Unmeasured: A matched restore and usage-ledger check has not been run here; existing source-control features are described in the product comparison.

**Multi-page & interactive applications**
- Documented: Agent-assisted application building and development workspace. Sources: https://replit.com/pricing.
- Unmeasured: An equivalent interactive workload has not been run in this evaluation; this does not imply an inability to build games.
- Unmeasured: No matched workload/device frame-time measurement is recorded here.

### v0

Overall score: not measured. 6 documented or recorded checklist conditions; 14 unmeasured; 0 explicitly unmet. This is evidence coverage, not product quality.

**Code ownership & export cost**
- Documented: Source can be exported and deployed outside Vercel. Sources: https://v0.app/docs/faqs.
- Documented: GitHub-backed branches, pull requests and CI workflows. Sources: https://v0.app/docs/deployments.
- Unmeasured: Plus is $30 per user per month at the checked published offer; one user is modeled. An equivalent complete source-access purchase has not been priced by a matched export test.
- Unmeasured: This specific source-access condition has not been established here; it is not marked unsupported.
- Documented: Exported code can move; reconfigure framework runtime and service integrations. Sources: https://v0.app/docs/faqs.

**Database, accounts & app AI**
- Documented: Database and API integrations; provision services as needed. Sources: https://v0.app/docs/faqs.
- Documented: Implement app authentication with the chosen provider. Sources: https://v0.app/docs/faqs.
- Unmeasured: App-user inference has not been established in this source set; builder AI alone is not treated as app AI.
- Unmeasured: An equivalent schema-backed SDK contract has not been inspected here.
- Unmeasured: This automation condition has not been established here; supported integrations may provide it.

**Targeted edit precision & speed**
- Unmeasured: Exact context-selection behavior needs a documented or observed matched test; repository access alone does not prove surgical precision.
- Unmeasured: Equivalent selected-element context has not been checked here.
- Unmeasured: No equivalent before/after edit-accuracy benchmark is recorded here.
- Unmeasured: No comparable end-to-end edit-timing distribution is recorded here.

**No-usage editing & recovery**
- Unmeasured: No-usage visual-edit terms have not been established here; absence of a recorded check is not a paid-only claim.
- Unmeasured: The same injected fault and actual customer usage ledger have not been inspected on this product.
- Unmeasured: A matched restore and usage-ledger check has not been run here; existing source-control features are described in the product comparison.

**Multi-page & interactive applications**
- Documented: AI UI and full-stack app development by Vercel. Sources: https://v0.app/docs/faqs.
- Unmeasured: An equivalent interactive workload has not been run in this evaluation; this does not imply an inability to build games.
- Unmeasured: No matched workload/device frame-time measurement is recorded here.

## Recorded Uvilno checks

- **database-persistence:** A code-authored integration fixture persisted a task through Uvilno.db and restored it after iframe reload; also checked on production.
- **transactional-automations:** The same fixture persisted an activity record and notification in the task transaction; also checked on production.
- **app-ai:** Two actual free-model requests returned text and a parsed JSON object through the generated-app SDK in local integration testing.
- **identity-isolation:** Guest, Member and Admin presentation changes left the real visitor identity and creator session unchanged; also checked on production.
- **source-lock:** Locked source returned HTTP 403; the owned context response contained metadata without record bodies or source; also checked on production.
- **bounded-static-repair:** An injected ReferenceError was reproduced in an isolated cloud browser and repaired in app.js by one free-model response, then rechecked; no customer usage-ledger entry was added.

The 50-case comparative generation protocol has not been run. The six first-party checks do not establish cross-vendor completion rates, quality scores or edit timings. [Download the protocol and recorded checks](https://uvilno.com/benchmarks-2026.csv).

## Managed database, automations and in-app AI

Uvilno combines schema-declared collections, the Uvilno.db CRUD SDK, an owner-facing App Data manager, transactional database automations and real Uvilno.ai.ask/json inference. App AI consumes the creator's finite allowance and has a 600-output-token response ceiling. These are Uvilno-hosted services, rather than an independent backend provisioned inside a source ZIP.

## Source access with a one-time project unlock

Uvilno offers a $4.99 one-time permanent source unlock for one project. Plus costs $10/month with two included project unlocks per monthly period; Pro costs $20/month with unlimited source unlocks during the subscription, and Ultra costs $59/month. Generation, hosting and third-party services are separate. Source export does not migrate managed customer data, authentication or payment infrastructure.

## Specific editing context without a speed guarantee

Uvilno accepts file/page mentions such as @index.html, collection mentions such as @db:tasks and an inspector-supplied @selected-element. A private repository inventory and bounded affected-file retrieval guide edits while preserving unrelated modules. Source ownership and verification checks still apply. No published cross-vendor test establishes a universal 4–8 second edit time.

## Usage and limited no-usage editing

Uvilno displays generation capacity from 100% remaining to 0%. Mapped direct text, link and color edits use no model allowance. A separately rate-limited, platform-funded repair path can reproduce and correct native static-project runtime faults before saving. This is not an unlimited free coding endpoint or a guarantee that every Next.js error is repaired without usage.

## Project-isolated visitor accounts and integrations

Configured Google, GitHub, Discord and email visitor authentication uses project-scoped identities separate from the creator's Uvilno session. GitHub import, push and pull require authorization and applicable source entitlement. Stripe Connect and Checkout currently operate in test workflows; live payments require a separately configured release. Preview personas test presentation and cannot grant real admin permissions.

## Interactive applications and browser games

Uvilno can produce modular, multi-page frontend applications and browser games using Canvas, WebGL and its game runtime. Actual frame rate depends on the generated workload, browser and hardware. Source export and UI tests do not by themselves demonstrate production readiness, security or 60 FPS for every generated application.