Update, September 29, 2026: Sonnet 5 API pricing is now $2 per million input tokens and $10 per million output tokens. The introductory price was made permanent in August 2026. Anthropic has since released Claude Sonnet 5.5. That release does not by itself tell you which model Lovable uses to build your project.
Anthropic released Claude Sonnet 5 on June 30, 2026, positioning it as a mid tier model with stronger agentic coding. If you build software by describing it rather than typing it, the practical questions are simple: what does it change, what does it cost, and does it affect the tools you already use?
Three different prices, not one
People often mix up three separate things.
- API pricing. What developers pay Anthropic per token when they call Sonnet 5 directly: now $2 per million input tokens and $10 per million output tokens.
- Claude subscription pricing. What you pay for a Claude plan in the Claude apps. That is a monthly subscription, not a per token price.
- Lovable credits. What you spend inside Lovable. Lovable credits are their own unit, and their cost does not follow Anthropic's API price list. See Lovable credits and usage.
A lower API price is good news for anyone building their own agents on the Anthropic API. It does not automatically lower your Lovable credit costs.
What Anthropic says changed
Anthropic's launch post describes better performance on agentic coding tasks, where the model reads a codebase, plans, edits several files and checks its work. For the exact benchmark names and figures, read the official announcement rather than secondhand summaries, since benchmark results are easy to misquote.
In practical terms, three behaviors matter most for vibe coders when evaluating any new coding model:
- Does it plan before editing, instead of jumping straight into files?
- Does it recognize when it is stuck, instead of repeating the same failed fix?
- Does it keep track of a larger codebase without editing the wrong function?
These are the questions to test in your own workflow. Your results will depend on your project, your prompts and the tool you use the model through.
Where you might encounter it
- Claude Code and the Claude apps. Anthropic's own tools are the most direct way to use Sonnet 5. See the Claude Code vs Cursor vs Copilot comparison for context.
- IDEs such as Cursor. If your editor lets you choose models, you can select Sonnet 5 and compare it on your own multi file changes.
- Your own agents. If you call the Anthropic API directly, the new $2 and $10 pricing applies to your token bill.
- Lovable. Lovable does not publish one fixed builder model for every account, and you do not choose the builder model yourself. Separately, your Lovable app can call AI models through the Lovable AI gateway for features like chat. Check the Lovable changelog for what is actually available.
Prompting habits that help with any modern model
- Give the constraint, not the solution. Describe the problem and the result you want. "Add a booking flow that saves to the database, sends a confirmation email, and shows a success screen" works better than a long list of files to touch.
- Ask for a plan first on anything non trivial. In Lovable, that is what Plan mode is for. Elsewhere, "before you write code, walk me through your plan in five bullets" does the same job.
- Stop rewarding loops. If the agent tries the same thing twice, cancel and give it new information.
- Keep a useful Knowledge Base. In Lovable, keep the product facts, constraints and instructions that actually help. Remove only what is outdated.
The honest limits
Sonnet 5 is a mid tier model. For very hard reasoning tasks, a top tier model may still do better. It is also not a security review. It will wire an insecure auth flow if you ask for one. The beginner mistakes guide still applies in full. Better models raise the ceiling of what one prompt can do; they do not remove the need to review what ships.
Should you change anything?
If you build on the Anthropic API, recalculate your costs with the new pricing. If you use Cursor or Claude Code, try Sonnet 5 or Sonnet 5.5 on a real task and compare. If you build on Lovable, nothing needs changing on your side; keep an eye on the changelog.
If you have never tried Claude directly, the free week of Claude is the easiest way to see the model outside a builder. Honest disclosure: it earns me usage credits if you subscribe.
The bigger picture
Model releases keep raising what a single prompt can do, and cheaper API pricing makes custom agents more affordable. Combined with features like Lovable's agent integrations, the distance between an idea and a running app keeps shrinking. If you have been meaning to start, open Lovable and build something small this week.
