Haiku 5.5 cuts token prices by 90% for prompts up to 100,000 tokens
Anthropic’s Haiku 5.5 charges one-tenth of Haiku 4.5’s standard token rates for prompts up to 100,000 tokens, with a higher tier beyond that boundary.
Anthropic released Haiku 5.5 on October 7 for repetitive work such as classification, extraction and summaries, with an adjustable effort setting.
The new prices make a bounded trial of smaller models more attractive for jobs that do not need the strongest reasoning available. That is a deployment opportunity, not evidence that every workload should move. [1]
For prompts up to 100,000 tokens, standard input costs $0.10 per million tokens and output $0.50, against Haiku 4.5’s $1 and $5. Above the threshold, those Haiku 5.5 rates become $0.50 and $2.50. The higher tier applies to the request; cached input also counts towards its prompt length. [1][2]
The difference between a 90% unit-price cut and Anthropic’s roughly 75% average running-cost claim matters. The company’s average accounts for its request mix and a changed tokenizer. Your bill depends on the tokens and effort needed to finish your own task. Independent practitioner Ryo Suzuki reports successful Bedrock calls and a retrieved price list, corroborating availability and the global rates, without establishing general performance. [1][3]
Our practical recommendation: replay a small, representative set of completed tasks and compare accepted results, total charges and elapsed time. Keep the prompt below the price boundary when that preserves the information needed; reducing context at the cost of correctness is a false saving.
Sources & details 4 sources
Discuss in your voice app
Take the reporting, visual explanations and original sources into a conversation. Copy the brief, paste it into your preferred assistant, then start its voice mode.
Paste the brief into your voice app to discuss this reporting.Revised 9 October 2026. Copy edited 9 October to remove commentary about graphic design; reported claims, qualifications and source dates preserved.
Reporting sources
- Anthropic · October 7 release
Original launch: standard rates, effort controls, request-mix estimate and narrow-task positioning.
- Claude Platform · Pricing
Living original documentation checked October 9: both tiers, cache accounting and request-level threshold.
- Ryo Suzuki · Bedrock calls and price retrieval
October 8 firsthand practitioner report: invocation output, CloudTrail records and retrieved price list; no broad performance evaluation.
- SiliconANGLE · October 7 reporting
Duncan Riley corroborates the launch and previous-model prices; performance claims remain attributed to Anthropic.
Images & figures
A fold changes the terms. AI-generated editorial illustration of Haiku’s request-size boundary; no price, speed or token count is plotted.
Euangelion · AI-generated editorial illustration · Source · Usage terms
Compare input and output rates against Haiku 4.5; expanded reading separates the two prompt tiers.
Euangelion · original explanatory graphic · Source · Usage terms · Download figure