Use Claude Haiku 5.5 for quick, high-volume jobs like sorting messages, summarising notes and pulling details out of documents, and Claude Sonnet 5.5 for everyday writing, analysis and well-scoped tasks that need more judgement. Both are on Claude's Free plan. On the API, Haiku costs from $0.10 per million input tokens and Sonnet $2, as of October 2026.
They're siblings in the same Claude 5.5 family, and Anthropic positions them very differently. Haiku is the speed and volume model; Sonnet is the balance of speed and intelligence. Here's the side-by-side from Anthropic's own model pages and docs, and how I'd split the work between them.
Sonnet 5.5 vs Haiku 5.5 side by side
| Claude Sonnet 5.5 | Claude Haiku 5.5 | |
|---|---|---|
| Anthropic's description | The best combination of speed and intelligence | For high-volume, latency-sensitive tasks such as classification, extraction and routing |
| Speed | Fast | Fastest |
| API input / output per million tokens (as of October 2026) | $2 / $10 | $0.10 / $0.50 for prompts up to 100K tokens; $0.50 / $2.50 above |
| Default effort | High | Medium |
| Context window | 1M tokens | 1M tokens |
| Max output | 128K tokens | 128K tokens |
| Knowledge cutoff | June 2026 | June 2026 |
| Retirement | Not sooner than 28 September 2027 | Not sooner than 7 October 2027 |
The price gap is the headline. For prompts under 100K tokens, Sonnet's input rate is 20 times Haiku's. That doesn't matter much in the Claude app, but it matters a lot if you're building a chatbot or running thousands of tasks through the API.
When to pick Haiku
Anthropic says to use Haiku 5.5 when speed and volume matter most. Its model page lists the jobs it's built for:
- High-volume work such as summarisation, classification and text generation.
- Real-time experiences: chat, voice agents, live support and in-app assistants.
- Sub-agent work, where a bigger model plans the job and hands Haiku small, well-defined pieces.
- Repetitive computer use, like form filling, data entry and moving information between apps.
- Focused coding: direct edits and small changes that need applying across many files.
When to pick Sonnet
Anthropic describes Sonnet 5.5 as a faster, lower-cost complement to Opus 5.5 that's strong at well-scoped everyday tasks across coding, agents and knowledge work. It also says Sonnet's writing is clearer and more direct, so drafts need less cleanup, and that it has a strong eye for design, adding polish to slides, diagrams and user flows. For a newsletter draft, a client proposal outline or an analysis of last month's sales, Sonnet is the safer pick.
A quick test to try both
Run the same prompt on each model and compare. Here's one for a small business:
Copy this prompt
Here are 20 customer emails. Sort each one into: refund request, delivery question, product question or other. Then give me a two-line summary of the most common problem.
Haiku should handle the sorting quickly and well; that's exactly its job. Then try a second prompt that needs judgement:
Copy this prompt
Using those emails, draft a short, warm reply I could send to customers about the most common problem, and suggest one change to my website that would stop it happening.
This is where Sonnet usually earns its place, because the reply needs tone and the suggestion needs reasoning. If Haiku's draft is good enough, you've found a job you can keep on the cheaper model.
How to switch between them
- Click the model name next to the send button in Claude.
- Choose Sonnet 5.5 or Haiku 5.5; click More models if you can't see one.
- Pick an effort level. Haiku 5.5 is the first Haiku with effort controls, so you can trade speed against depth.
- Change model mid-chat any time. Anthropic says the change applies from Claude's next reply.
Where you can use each model
Both models are on Claude.ai on the web, iOS and Android, and Anthropic says Haiku 5.5 is also in Claude Code. Developers can call them on the Claude Platform as claude-sonnet-5-5 and claude-haiku-5-5, and through Amazon Web Services, Google Cloud and Microsoft Foundry. Prompt caching and batch processing cut the API cost of either model further.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Two pricing details are easy to miss. Haiku's low rate applies to prompts up to 100K tokens; longer prompts move to $0.50 input and $2.50 output per million tokens, as of October 2026. And Sonnet offers US-only inference at 1.1x its normal price for workloads that must run in the US. Anthropic also suggests pairing them with bigger models: a more capable model such as Opus or Fable plans the work and hands sub-tasks to Haiku, so you can run many small agents in parallel without paying top rates for every step.
Running lots of small jobs through Claude? If you'd rather have them set up for you, read how to hire an AI automation agency.
Thinking settings differ on the API
In the Claude app, thinking can't be switched off on either model. On the Claude API they behave differently: Haiku 5.5 lets you turn thinking off at effort levels high and below, but returns an error if you try at extra high or max. Sonnet 5.5 returns an error if you set thinking to disabled; Anthropic says to use the between_tools setting instead, so it only thinks between tool calls. If you're moving code from one model to the other, check this first.
For the full Haiku walkthrough, see how to use Claude Haiku 5.5, and for the other side of the family, Claude Opus vs Sonnet.
My verdict
If you're chatting in the Claude app, leave Sonnet as your default and switch to Haiku when you've got a pile of quick, repetitive jobs. If you're building something on the API, start with Haiku and only move a task up to Sonnet when Haiku's answers fall short. Either way, you'll save usage without losing quality where it counts.
Effort settings change how far your allowance goes, so read Claude extended thinking as well.
Frequently asked questions
Is Claude Haiku better than Sonnet?
Not overall. Haiku 5.5 is faster and cheaper; Sonnet 5.5 is stronger at writing, analysis and tasks needing judgement.
Is Claude Haiku free?
Yes. Anthropic says Free, Pro, Max, Team and Enterprise users can select Haiku 5.5 on Claude.ai.
How much cheaper is Haiku than Sonnet?
As of October 2026 on the API, Haiku 5.5 is $0.10 per million input tokens for prompts up to 100K tokens; Sonnet 5.5 is $2.
Which Claude model is fastest?
Anthropic's models overview lists Haiku 5.5 as the fastest current Claude model.
Can I use Haiku 5.5 in Claude Code?
Yes. Anthropic says Claude Haiku 5.5 is also available in Claude Code.
