Claude Haiku 5.5
- Input / 1M tokens
- $0.10
- Output / 1M tokens
- $0.50
API models · Side by side
Claude Haiku 5.5 vs Claude Sonnet 5.5
Choosing between Haiku and Sonnet is largely a question of how much work each request needs to do. A short classification job and a multi-file coding change should not automatically use the same budget.
Claude Sonnet 5.5
- Input / 1M tokens
- $2.00
- Output / 1M tokens
- $10.00
The short answer
Haiku has the lower base token rates. Both accept text, images and files, with the same context and maximum output allowances. The important cost wrinkle is Haiku’s higher pricing tier for long prompts.
Compare the details
Compare context, supported inputs and usage costs. Highlighted rows show a difference.
| Feature | Claude Haiku 5.5 | Claude Sonnet 5.5 |
|---|---|---|
| Input formats | file, image, text | file, image, text |
| Output formats | text | text |
| Context / position limit | 1,000,000 | 1,000,000 |
| Maximum output tokens | 128,000 | 128,000 |
| Developer features | Reasoning output, Completion limit, Output limit, Reasoning controls, Thinking effort, Response format, Stop sequences, Structured output, Tool selection, Tool calling, Response length | Reasoning output, Completion limit, Output limit, Reasoning controls, Thinking effort, Response format, Stop sequences, Structured output, Tool selection, Tool calling, Response length |
| Cached input / 1M tokens | $0.01 | $0.10 |
| Cache write / 1M tokens | $0.13 | $2.50 |
| USD input / 1M tokens | $0.10 | $2.00 |
| USD output / 1M tokens | $0.50 | $10.00 |
| Model identifier | anthropic/claude-haiku-5.5 | anthropic/claude-sonnet-5.5 |
API pricing
Prices are in USD per million tokens. Input is what you send; output is what the model generates. Claude and ChatGPT subscriptions are billed separately.
Claude Haiku 5.5: long-prompt rates
- Long-prompt threshold: 100,000 prompt tokens. Input: $0.50 / 1M. Output: $2.50 / 1M.
Tools, images and additional reasoning can add charges. The examples below cover input and output tokens only.
Compare local modelsWhat those prices mean in practice
Each example uses the same token budget for both models. These are calculations, not measured workloads. Cache discounts, media, tools and extra reasoning are excluded.
| Workload | Claude Haiku 5.5 | Claude Sonnet 5.5 |
|---|---|---|
| 1,000 short requestsPer request: 2,000 input + 500 output tokens | $0.45 | $9.00 |
| 100 document summariesPer request: 20,000 input + 1,000 output tokens | $0.25 | $5.00 |
| 10 long-document requestsPer request: 300,000 input + 2,000 output tokens | $1.55 | $6.20 |
Estimate = requests × (input tokens × input price + output tokens × output price) ÷ 1,000,000. Long-prompt rates are applied where the pricing data provides them. Real requests can use different amounts of output.
Beyond the price tag
Where Haiku earns a place
Consider Haiku for work with a clear, checkable answer: assigning a support ticket category, extracting fields from an invoice or condensing a short update. Define the allowed categories or fields in the prompt and validate the response before it reaches another system. A lower token bill is most useful when the result can be checked cheaply. Retrying an ambiguous task repeatedly can erase part of the saving.
When to keep Sonnet in the workflow
A coding request that changes several files or a document task that must reconcile conflicting passages needs a different acceptance check. Keep Sonnet as a candidate for that work and decide using representative examples from your project. Compare whether the result meets your requirements, how much editing it needs and how many requests it takes. The model names and prices do not establish a universal quality ranking.
Long prompts change the budget
Haiku’s ordinary rate and long-prompt rate are different. A conversation can enter the higher tier as documents and earlier messages accumulate, even if the newest message is short. The worked examples apply the pricing tiers in the current data. For repeated document work, send only the relevant material or use eligible prompt caching; do not assume a million-token capacity means a million-token request has the base price.
A practical two-model arrangement
An application can use Haiku for extraction or routing and send selected work to Sonnet. Make the handoff explicit: pass the original material when the second model needs to verify a claim, rather than only an unchecked summary. Record the cost of both calls and any retries. This arrangement is useful only when the split reduces total work without losing information needed for the final answer.
Common questions
Does Haiku have a smaller context window?
Not for these two versions in the current comparison. Both allow one million context tokens. Their pricing rules differ, so the same long prompt can still have a different bill.
Can I replace Sonnet with Haiku without changing anything?
Keep the same response contracts, but check actual outputs, tool arguments and reasoning settings. A compatible input format does not guarantee identical behavior in an existing application.