# Anthropic Launches Haiku 5.5 at Up to 90% Lower Token Prices

Published 2026-10-07 · AI.info Team · AI.info

> Anthropic released Claude Haiku 5.5 on October 7, 2026, with sharply lower per-token prices than Haiku 4.5. The model targets high-volume tasks, while its updated tokenizer and weaker coding benchmark results qualify the savings and scope.

[Anthropic](https://www.anthropic.com/claude-haiku-5-5) is pricing Claude Haiku 5.5 at one-tenth the previous model’s rates for many requests, but the lower token price does not translate directly into the same percentage of savings on every task. The new model’s tokenizer uses slightly more tokens to complete a task, and requests above 100,000 prompt tokens receive a smaller discount. Haiku 5.5, announced October 7, is aimed at high-volume work where those trade-offs matter: summaries, classification, database queries and subagent tasks.

## Haiku 5.5 cuts prices, with a long-prompt tier

For prompts up to 100,000 tokens, Anthropic charges $0.10 per million input tokens and $0.50 per million output tokens. Haiku 4.5 costs $1 and $5, respectively, at those rates. Above 100,000 prompt tokens, Haiku 5.5 costs $0.50 per million input tokens and $2.50 per million output tokens—a 50% reduction against Haiku 4.5, rather than 90%.

About 90% of requests to its previous Haiku model fell into the lower-priced prompt tier, the company says. Across usage, Haiku 5.5 costs about 75% less to run on average, Anthropic estimates. That estimate accounts for the updated tokenizer, which uses slightly more tokens per task; the company does not publish a single guaranteed reduction for every workload.

## Anthropic positions the model below Sonnet

Haiku 5.5 is designed for work that can be divided into quick, repetitive tasks, or assigned to a smaller model while a larger one handles the main job. The model is best suited to compaction, summarization and subagent work, Anthropic says, while Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding. That distinction puts the new model’s lower price alongside a narrower intended role, rather than presenting it as a replacement for Anthropic’s larger models.

Anthropic’s own benchmark table reflects that gap. On Terminal-Bench 4.0, which tests multi-step command-line tasks, Haiku 5.5 scores 39.2%, compared with 70.6% for Sonnet 5.5. Those are company-published results, not an independent evaluation. Haiku 5.5 is also its fastest model at standard speed, Anthropic says, while noting that it runs more slowly than Opus models in Fast Mode.

## Early customer tests point to speed gains

Customers cited in the announcement describe benefits on narrower, production-oriented workloads. Aaron Vinh, a staff software engineer at Asana, said the company saw “over a 30% reduction in latency for task completions” in its AI Teammates evaluation, alongside faster inference per agent turn. Anthropic does not specify in the post which previous model Asana used for that comparison, so the result cannot be read as a head-to-head against Haiku 4.5.

[HubSpot](https://ai.info/companies/hubspot)’s Ze’ev Klapow, a distinguished software engineer, said Haiku 5.5 scored 92.8% on the company’s CRM evaluation, averaged over three runs. That is a HubSpot test result, not a general benchmark. The announcement also cites a 400-query [AlphaSense](https://ai.info/companies/alphasense) test, where the company says Haiku 5.5 improved on Haiku 4.5 from 0.76 to 0.84.

## Availability extends across Anthropic’s platforms

Haiku 5.5 is available now through the Claude Platform, [Amazon Web Services](https://ai.info/companies/amazon-web-services), [Google Cloud](https://ai.info/companies/google-cloud) and Microsoft Azure, under the API model ID `claude-haiku-5-5`. The release also brings a separate price change for Sonnet 5.5: cache reads drop by half, to $0.10 per million tokens from $0.20. That change will reduce Sonnet 5.5’s cost by about 20% on most agentic tasks, Anthropic estimates.

For Haiku 5.5, the practical price difference depends on prompt length, token use and the task’s fit for a smaller model. Its lowest rates apply to the prompt tier Anthropic says covers most previous Haiku requests; long prompts face the higher tier, and complex coding remains a stated use case for Sonnet or Opus instead.

## Sources

1. Original source [Anthropic](https://www.anthropic.com/claude-haiku-5-5) anthropic.com

## Companies in this story

- [Anthropic](https://ai.info/companies/anthropic)

## Related stories

- [McDonald’s AI Pricing Engine Estimates What Diners Will Pay](https://ai.info/news/mcdonalds-ai-pricing-engine-diners-willingness-to-pay)
- [Anthropic Finds Small Models Near AI-Worm Self-Replication](https://ai.info/news/anthropic-small-models-ai-worm-self-replication)
- [Meta Introduces Muse Wonder Models for Agents and Coding](https://ai.info/news/meta-launches-muse-wonder-models)
- [Google Opens Gemini 4 Argon to Cybersecurity Defenders](https://ai.info/news/google-gemini-4-argon-cybersecurity-defenders)
- [Meta Releases 816B-Parameter Muse Spark Open Weights](https://ai.info/news/meta-releases-muse-spark-open-weights)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/anthropic-launches-haiku-5-5-lower-prices
