New in Bolt: Use Claude Opus & Haiku

Starting today, Bolt users can switch between Claude’s three 4.5 models (Haiku, Sonnet, and Opus) directly in the agent sidebar. 

Each Claude model has different strengths: Haiku works fast, Opus has deeper reasoning, and Sonnet (the default model in Bolt) strikes a balance between the two. 

It’s hard to choose which model is best for your project. Sometimes you need speed, other times precision. So we’re making all three available to you right in the editor. Just toggle between models as your needs change. 

This update introduces a new dimension of control to AI-assisted development. You can optimize for speed, complexity or balance depending on the task at hand. 

One agent. Three models. Optimized for any task.

Different tasks require different skillsets. If you’re iterating fast on tiny UI changes, a model built for speed will do the trick. But if you’re designing a core system from scratch, you’ll want something capable of deeper reasoning. 

Now, you can match the right Claude model to the job without switching agents or losing context. 

Model breakdown

Here’s a look at where each model shines.

Haiku 4.5: Fastest responses, lowest token usage. 

Great for quick edits, layout tweaks, translations, boilerplate. Use it for:

• Styling tweaks

• Text edits & translations

• Simple form pages (TOS, login, etc.)

• Quick prototyping & ideas

• Generating test data

  
  

You’ll get results almost instantly and save on tokens.  It makes sense for fixing broken buttons, styling forms, and other straightforward tasks. 

“Just fixed copy, tweaked a layout, and updated a form — all in 90 seconds. Haiku is my new go-to for speed.” - Real Discord user. 

Sonnet 4.5 (your current default): Balanced performance. 

Sonnet strikes a balance between speed, capability, and reliability. Ideal for building features, debugging, and full-stack work. Use it for:

• Building features

• Backend + database design

• UI work with finesse

• Integrations & debugging
  
  

This is the Claude model you’ve already been using in Bolt. It’s still the best all-around performer for full-stack work. Perfect for adding a Stripe checkout flow, building a dashboard, or other tasks of moderate complexity. 

Opus 4.5: Deep reasoning. (Most capable model).

Best at complex features, security, and architecture decisions. Use it for: 

• Big refactors

• Multi-step debugging

• Architecture decisions

• Advanced performance optimization
  
  

Opus is more context aware and generates fewer hallucinations. It uses more tokens, but it’s worth it for harder problems and more complex builds. Bring in the big guns when you’re refactoring a monolith or designing a secure authentication flow, for example. 

How to switch between Claude models

Just click into the Claude agent in your workspace. 

You’ll see a model selector in the sidebar dropdown. Pick the one you want to use.

Models that make sense

This update ensures you have the right tool for the job. Nothing more and nothing less. Now, you don’t have to waste tokens on small fixes. But when you need the extra horsepower for a complex project, you’ll have it on hand.

Try it out now.

FAQ

What happens if I stay on Sonnet?

Nothing changes. Sonnet is still your default and it’s a great choice for most dev work.

Will Haiku and Opus cost more?

All models use tokens. Haiku is the most efficient. Opus uses more tokens for better reasoning. You choose based on your needs.

Can I switch mid-thread?

Not yet. We recommend using different threads for different models/tasks.

Why not use Codex or Gemini 3 Pro?

We evaluate every major model and pick the best for coding. Right now, Claude 4.5 family outperforms the rest in real-world use. When something better comes along, we’ll bring it to Bolt.