Free tools Windows power users keep installed
One-click scans. No signup required.
As of October 7, 2026, Anthropic lists Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. For prompts over 100,000 tokens, the rates rise to $0.50 input and $2.50 output per million. Haiku 5.5 has a 1-million-token API context window and a regular maximum output of 128,000 tokens. Sonnet 5.5 costs more per token—$2 input and $10 output per million—with the same listed context and output ceilings. Those model limits are different from the API throughput and spend limits assigned to an organization.
How much does Claude Haiku 5.5 cost?
Anthropic’s published API rates are in U.S. dollars per million tokens. Haiku 5.5 has two price tiers based on prompt length: requests with prompts up to 100,000 tokens use the lower rates; prompts over 100,000 tokens use the higher rates. Check the Anthropic pricing page before budgeting, because prices and billing terms can change.
As an Amazon Associate I earn from qualifying purchases.
| Haiku 5.5 prompt length | Input price per 1M tokens | Output price per 1M tokens |
|---|---|---|
| Up to 100,000 tokens | $0.10 | $0.50 |
| Over 100,000 tokens | $0.50 | $2.50 |
For a simple estimate, calculate (input tokens × input rate + output tokens × output rate) ÷ 1,000,000, using the rate tier that applies to the prompt. For example, a 100,000-token prompt and 20,000-token completion at the lower rates would cost about $0.02 in base token charges: $0.01 for input and $0.01 for output. This estimate excludes caching, tools, and any charges imposed by the service provider handling the request.
How does Haiku compare with Sonnet and other small models?
The comparison below uses published API token rates. The Sonnet 5.5 rate and capacity figures are from Anthropic’s pricing and model information; its overview describes it as a balance of speed and intelligence, which is vendor positioning rather than an independent quality assessment. Google’s Gemini figure is only a price reference from the cited Google Cloud table, not a like-for-like comparison of capability or total cost.
#1 Best Overall
| Model | Input / output price per 1M tokens | Context / maximum output | What the figures establish |
|---|---|---|---|
| Claude Haiku 5.5 | Up to 100K prompt: $0.10 / $0.50; over 100K prompt: $0.50 / $2.50 | 1M / 128K | Anthropic’s current Haiku listing; the prompt-length tier matters. |
| Claude Sonnet 5.5 | $2 / $10 | 1M / 128K | Higher published per-token rates than Haiku 5.5; the same listed context and output ceilings. |
| Claude Haiku 4.5 | $1 / $5 | 200K / 64K | Previous-generation model marked legacy by Anthropic. Its overview says retirement will not occur sooner than October 15, 2026; check the current lifecycle status before planning a migration. |
| Gemini 3 Flash Preview (Google Cloud pricing reference) | $0.25 input / $1.50 text output | Not stated in the cited price row | A limited cross-provider price point; the cited row does not establish comparable limits, availability, quality, or total cost. |
Haiku 5.5’s lower starting rates make its published token price substantially below Sonnet 5.5’s, but long prompts move Haiku to its higher tier. Token prices alone do not determine which model costs less for a particular job: prompt and completion sizes, cache use, tools, billing route, and the work’s requirements all matter. The cited Google Cloud price does not support a quality ranking against either Claude model.
What do Haiku’s context and output limits mean?
- Context window: The Haiku 5.5 API overview lists 1 million tokens. This is the size of the request or conversation context the model can handle, not a monthly usage allowance.
- Maximum output: A standard API request can generate up to 128,000 tokens. The overview separately lists a 300,000-token maximum for Message Batches API in beta with a specified beta header; that conditional batch capability is not the regular request limit.
- Throughput and spend: These are organization-level API limits, described in the next section. A large context window does not mean an account can send unlimited requests or tokens.
Anthropic positions Haiku 5.5 for “high-volume, latency-sensitive tasks such as classification, extraction, and routing.” That is the provider’s intended-use description, not an independent test result. The overview also lists Claude availability through the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry; limits and billing can differ by route. See the Haiku 5.5 model overview for its model-specific details.
What API rate limits and spend caps apply?
Anthropic documents standard Haiku 5.5 limits for Start, Build, and Scale tiers. RPM means requests per minute; input and output token limits are separate per-minute ceilings. The monthly spend caps shown are documented tier caps, not a guarantee that every organization has identical assigned limits. Anthropic says accounts may have lower evaluation limits or customized limits; check the assigned figures in Claude Console.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →| Organization tier | Requests per minute | Input tokens per minute | Output tokens per minute | Documented monthly spend cap |
|---|---|---|---|---|
| Start | 1,000 | 2 million | 400,000 | $500 |
| Build | 5,000 | 5 million | 1 million | $1,000 |
| Scale | 10,000 | 10 million | 2 million | $200,000 |
| Custom | Arranged with the account team | Arranged with the account team | Arranged with the account team | Arranged with the account team |
These are the standard limits published by Anthropic, not a substitute for checking your organization’s actual quota. Current tier details and account-specific limits are on the Anthropic API rate limits page.
Can caching or batch processing change the cost?
Yes. For prompts up to 100,000 tokens, Anthropic’s Haiku 5.5 overview lists cache-write rates of $0.125 per million tokens for a 5-minute cache and $0.20 for a 1-hour cache; cache reads are $0.01 per million tokens. For prompts over 100,000 tokens, the corresponding cache-write rates are $0.625 and $1 per million tokens, and cache reads are $0.05. Anthropic also says batch processing discounts input and output token rates by 50%. These items change the estimate, so consult the complete Haiku 5.5 pricing details and pricing page for the exact billing conditions that apply.
Are Claude app limits the same as API limits?
No. Anthropic’s Help Center lists Haiku 5.5 context at 1 million tokens in Claude chat and 500,000 tokens in Cowork. Those hosted-product context figures are separate from API billing and organization quotas. The Help Center also says that on paid plans, automatic context management can summarize earlier parts of long conversations when code execution is enabled; longer conversations using that feature consume more of the plan’s usage limit. See Anthropic’s context-window guidance for paid Claude plans.
Rank #4
Which model should you choose?
Use the published figures to screen options, then compare them against the actual workload rather than treating a lower token rate as proof of better value.
- Consider Haiku 5.5 when its price and listed capacity fit the task, particularly if the prompt stays within 100,000 tokens and the workflow benefits from high-volume processing. Anthropic’s use-case description is vendor guidance, not a guarantee of performance.
- Consider Sonnet 5.5 when you need to evaluate Anthropic’s higher-priced alternative. Its 1M context and 128K output match Haiku 5.5’s listed ceilings, but the cited material does not establish a universal quality advantage for a given task.
- Evaluate Gemini 3 Flash Preview separately if Google Cloud is a relevant billing route. The cited price row gives a rate reference, but does not establish enough about context, output limits, or equivalent workload performance for a full comparison.
- Before committing, compare prompt and completion token volumes, prompt-length pricing tiers, context and output needs, throughput and spend quotas, caching or batch use, supported features for the task, and the provider route and geography available to your organization.
For a production decision, estimate charges from real token counts and confirm account-assigned limits and provider billing terms. Published model ceilings do not establish the latency, quality, or cost of an untested workload.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
