v0

You will like v0 if your stack is already Next.js and shadcn, because the output matches the code you would have written and lands in a GitHub working branch and a Vercel deploy without ceremony. That is the daily trait that decides it: it speaks your framework's dialect. Where it hurts: the models are Vercel's own composites you cannot swap, so when the output is wrong there is no better model to reach for.

Pick it for Vercel shops that want UI scaffolding from a prompt. Pick Lovable if you want a backend included, and Bolt if you are not married to Next.js.

The worst thing is the pricing structure. Overage is per token, up to $50 per million output tokens on Max Fast, which is frontier pricing for a model whose composition the vendor does not disclose, and the subscription tiers include exactly as many credits as they cost, so the plan buys access rather than usage. A long session on the top model is a bill with no ceiling but your own.

Pin the model tier and set a credit cap before a team touches it. What it does right: a browser tool opens the built app, so the agent sees the page it produced rather than guessing.

Edits and commands execute in a sandbox with permission modes Ask, Auto and Full, so the blast radius is a setting rather than a hope, and the default requires consent. That is the legible half. The models are the unprincipled half: "v0 Mini, Pro, Max and Max Fast" are composites with no statement of the underlying models or routing, so a result cannot be attributed to anything and cannot be reproduced when the composition changes silently.

No benchmark exists, and the Platform API is the only surface fully specified. The observation: the cheapest way to learn what a v0 model is would be to ask it.

Vercel does not need v0 to make money on v0: every generated app deploys to Vercel, and Next.js is distribution competitors would pay for, so the free plan's $5 of monthly credits is a subsidy that pays for itself in hosting. The price can rise without churn because churn lands on the hosting bill, and hosting is where the margin lives.

No acquirer question; v0 is its parent's funnel, and the parent is not for sale. The pivot, if any, is toward the enterprise seat with governance attached. Position: long the parent, indifferent to the tool's own price.

The demo is a UI from a prompt. Procurement: the training opt-out is a Business feature at $100 per user, so $6,000 a month for sixty seats is the real price of a clean data answer, and SAML lives on a custom Enterprise contract above that. Throughput gain for a backend engineer is zero, so most of those seats would buy nothing.

CI fit is a headless API, which the frontend team could use. Approved with conditions: Business tier, frontend seats only, a hard credit cap, and the opt-out verified in the account settings before the first prompt.

The models are Vercel composites I cannot replace: no bring-your-own-key, no local models, no source. What I can bend: an MCP server at v0.app/api/mcp as a plain url in my mcpServers config, so my own agent can drive v0 as a tool, npm install v0-sdk for scripting, and a Platform API underneath both. That is a better API surface than most closed builders offer.

A fork is impossible, an exit is easy, and the tool is honest about which. Grudging respect for the MCP server; a closed tool that speaks the protocol is at least a closed tool I can orchestrate.