"AI coding assistant" now covers products that do quite different jobs. Some suggest the next line while you type. Some take a whole task, work in the background and come back with a pull request. Others build and host an entire web app from a plain-language description, for people who may never read the code. Choosing well starts with knowing which of these you need.
This guide splits the Coding category into three types, sets out the criteria that separate them, and compares eight tools on the facts their vendors publish.
How we put this together
This is an evidence-based guide, not a hands-on review. Every product fact comes from that tool's listing in our directory, which is compiled from the vendor's own website, documentation and pricing page. Each listing names its sources and the date we read each one. We don't score or rank tools and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure. Coding plans change often, so confirm the current terms with the vendor.
Three kinds of AI coding tool
1. Assistants inside your editor. These work where developers already write code. GitHub Copilot offers inline completion, next edit suggestions, chat and an agent mode, with support for VS Code, Visual Studio, JetBrains IDEs, Xcode, Neovim and Eclipse. Cursor is a code editor built around AI, with agent and plan modes, targeted inline edits and a command-line agent.
2. Coding agents that take a whole task. Several products now run multi-step tasks with less supervision. Copilot has a cloud agent that works in the background and opens pull requests. Cursor runs cloud agents in parallel. Claude includes Claude Code on its paid plans. ChatGPT gives access to OpenAI's Codex coding agent. Mistral Vibe (formerly Le Chat) includes a coding agent available on the web, from a CLI and through IDE integrations.
3. Prompt-to-app builders. These generate a working app from a description and usually host it too. Replit's Agent writes the code, sets up a database and authentication, and publishes on Replit's hosting. Lovable generates full-stack web apps and syncs projects to GitHub, GitLab or Bitbucket. Bolt.new builds sites and apps from chat, can start from a Figma file or GitHub repository, and hosts them on Bolt Cloud. v0 turns prompts, wireframes or mockups into interfaces and full-stack apps using Next.js, Tailwind CSS and shadcn/ui, with GitHub sync and one-click deployment to Vercel.
Decision criteria
Who will read and maintain the code? If developers will own it, start with editor assistants and agents that work in your repository. If nobody on the team reads code, an app builder is the closer fit, but plan how you would move the project later.
Where will the code live? Lovable syncs with GitHub, GitLab and Bitbucket; v0 offers GitHub sync and pull requests; Bolt.new imports from GitHub; Replit includes its own hosting, database and authentication. Ask how you get a complete copy of your code and data out before you build anything important.
Does it fit your editor and review process? Copilot lists six supported editors plus a CLI and GitHub.com. Cursor is its own editor, with a CLI and Slack and GitHub integrations. Both list AI pull request review: Copilot code review and Cursor's Bugbot.
How much model choice do you need? Copilot lists a choice of several language models. Cursor lists models from several providers, and its listing notes that third-party model usage is billed separately. Bolt.new routes tasks to models automatically, but its free plan doesn't include the newest models.
What admin controls does your organization require? Copilot's Business and Enterprise plans add policy controls and license management. Replit lists SSO/SAML and admin controls for enterprises. Lovable lists SSO, SCIM and audit logs for organizations.
How is usage metered? Most tools combine a monthly price with a credit, request or message allowance, so a low headline price can still run out quickly. The table below shows the stated free-plan limits.
Eight tools compared on published facts
| Tool | Type | Pricing | Stated limits |
|---|---|---|---|
| GitHub Copilot | Editor assistant and cloud agent | Freemium, paid from $10/mo (checked 23 Oct 2026) | Free plan: 2,000 inline suggestions a month, no cloud agent or code review; extra usage beyond included credits must be bought |
| Cursor | AI code editor and agents | Freemium, paid from $8/mo (checked 24 Oct 2026) | Free plan has limited agent and chat requests; third-party model usage billed separately |
| Replit | App builder with hosting | Freemium, paid from $18/mo (checked 24 Oct 2026) | Free Starter plan allows one published project |
| Lovable | Full-stack app builder | Freemium; shared credit balance rather than seats | Free plan: 5 build credits a day, up to 30 a month; credits are not refundable |
| Bolt.new | App and site builder with hosting | Freemium, paid from $18/mo (checked 24 Oct 2026) | Free plan: Bolt branding on sites, 300K daily token limit, no newest models |
| v0 | UI and app builder | Freemium, paid from $30/mo (checked 24 Oct 2026) | Free plan: 7 messages a day |
| Claude | General assistant; Claude Code on paid plans | Freemium, paid from $17/mo (checked 24 Oct 2026) | Usage limits on every plan, lowest on the free plan |
| Mistral Vibe | General assistant with a coding agent | Freemium | Free plan has limited messages and web searches |
Starting prices are the lowest paid price recorded in each listing on the date shown, and several vendors note that their prices exclude tax. Check the billing period and any usage-based charges before you compare.
When not to rely on an AI coding tool
- Without a review step. Generated code needs the same review, automated checks and security scanning as code written by a person. An agent that opens pull requests saves typing, not review.
- On code your policies haven't cleared. If your organization restricts where source code may be sent, check the vendor's data terms and your own policy before connecting a repository.
- When the goal is learning. Accepting suggestions you don't understand undermines that goal. Asking the assistant to explain unfamiliar code, a use case Copilot lists, serves it better.
- For large existing systems built with an app builder. Prompt-to-app tools are oriented to starting new projects. For a mature codebase, editor assistants and repository-aware agents are the closer match.
A one-week trial plan
- Pick one real, small task, such as a bug fix, a small feature or an internal page.
- Try two tools of the same type on their free plans so the comparison is fair.
- Review the output as you would a colleague's pull request, and run your existing test suite and linters on it.
- Record where you hit a limit: messages, credits, suggestions or published projects.
- Check data and admin controls with whoever owns security before a team-wide rollout.
Questions to ask before paying
- Which editors, repositories and CI systems does it support, and which are paid-only?
- What is metered (requests, credits, tokens, messages), and what happens when the allowance runs out?
- Are third-party models billed separately?
- Can we export all code and data, and keep running the app without the vendor's hosting?
- Is our code used to train models, and can an administrator control that?
- Does the listed price include tax, and is it per user or shared across the workspace?
Related pages
- Browse all Coding tools and general Chat assistants with coding features.
- Search the directory for coding agent or app builder.
