Haiku 5.5 cuts API prices for small teams

Anthropic released Claude Haiku 5.5 on October 7, with a substantial price cut for the short, repeated jobs that can add up across a business. For a small team, the opportunity is to test more routine work: sorting support tickets, extracting fields from documents or making first-pass summaries before someone reads the originals. Sources: Introducing Claude Haiku 5.5, Claude Haiku

The price has a boundary

On Anthropic’s platform, prompts of up to and including 100,000 tokens cost $0.10 per million input tokens and $0.50 per million output tokens. Above that prompt length, the rates rise to $0.50 and $2.50. Haiku 4.5’s base rates were $1 and $5. These are US-dollar API charges, measured in tokens, rather than a price for a chat subscription. Sources: Claude pricing, Pricing

That is a 90% unit-price reduction for the shorter tier and 50% for the longer one. Anthropic’s separate estimate of about 75% lower average run costs reflects its previous request mix and changes in token consumption. It is a vendor estimate, not a savings guarantee for a particular app. Sources: Introducing Claude Haiku 5.5

A simple illustration: 10,000 calls billed at 2,000 input tokens and 500 output tokens each would incur $4.50 in base model charges at the lower rates, against $45 at Haiku 4.5 rates. This compares equal billable token counts. It excludes retries, tools, storage and other operating costs; it does not predict the bill for processing the same text. Sources: Claude pricing

Recount before switching

Anthropic’s migration guide says the same input text produces approximately 30% more tokens with Haiku 5.5’s newer tokenizer, depending on the content. A prompt that used to fit below a budget may cross it after migration. Teams should count their real prompts with the new model, check output limits and update thinking settings rather than only changing the model name. Sources: Claude Haiku 5.5 migration guide

Caching and batching offer further options. At the shorter tier, cache reads are $0.01 per million tokens; five-minute cache writes are $0.125 and one-hour writes $0.20. The longer-tier equivalents are $0.05, $0.625 and $1. The asynchronous Batch API halves standard input and output prices. Cloud endpoint choices and paid tools can add charges, so these figures are not an all-in quotation. Sources: Pricing, Claude Haiku 5.5 overview

Start with a narrow job

A sensible pilot has a clear output and an easy way to catch mistakes. For example, label tickets against a fixed list, extract an invoice reference with a link back to the source, or produce a short summary for human review. Keep uncertain cases visible and route them to a person or a stronger model. Measure accepted results, latency and retries alongside spending.

Haiku 5.5 supports a one-million-token context window and adjustable effort, with medium as the default. Availability includes the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry; features vary by platform. Anthropic’s launch table reports 39.2% on Terminal-Bench 4.0 against Sonnet 5.5’s 70.6%. Those are vendor-reported benchmark results, not Bright testing; headline scores do not establish performance at the default setting. They support retaining stronger models for demanding multi-step work; effort settings also affect the cost-quality trade-off. Sources: Claude Haiku 5.5 overview, Introducing Claude Haiku 5.5, Claude in Microsoft Foundry

API credits have separate rules

Anthropic also lists Haiku 5.5 for users across Free, Pro, Max, Team and Enterprise plans. New monthly API credits roll out over several days for Max and Team subscribers: $100 for Max 5x, $200 for Max 20x, and a Team pool capped at $500, based on seat types. New subscribers must wait seven days to claim. Unused credits expire each billing cycle. They do not increase Claude or Claude Code plan limits, cover interactive Claude Code or apply to partner-cloud usage. Free, Pro and Enterprise plans are ineligible for these credits. Sources: Claude Haiku, Monthly API credits for Max and Team plans

For founders, the practical test is whether a bounded workflow becomes affordable enough to use regularly while keeping quality checks in place. Compare the total cost of a reviewed, usable result before expanding it to more consequential work.

Screenshot

Anthropic’s Haiku 5.5 API pricing changes for prompts above 100,000 tokens. Rates shown are per million tokens, with batch processing off. Source: Anthropic (Claude.com). Screenshot captured for Bright. Official source. Standard API pricing; cache writes shown use a five-minute TTL. Limited accompanying editorial use; rights remain with the source owner. No express redistribution or onward syndication license.