---
id: "lib-2026-10-07-introducing-claude-haiku-5-5"
type: "library"
title: "Introducing Claude Haiku 5.5"
lang: "EN"
org: "Anthropic"
source_type: "Release"
date: "2026-10-07"
week: "2026-W41"
url: "https://nosov.ai/library/lib-2026-10-07-introducing-claude-haiku-5-5/"
original_url: "https://www.anthropic.com/claude-haiku-5-5"
tier: "C"
tier_name: "Weak signal"
tier_reason: "Vendor product announcement from Anthropic; benchmark and customer figures are self-reported or supplied by launch partners."
summary_basis: "full-text"
original_minutes: "7"
tags: ["ai technology", "model release", "token cost"]
---

# Introducing Claude Haiku 5.5

Tier C (Weak signal): Vendor product announcement from Anthropic; benchmark and customer figures are self-reported or supplied by launch partners.

Summary based on the full text.

## In short

Anthropic's launch post for Claude Haiku 5.5, published 7 October 2026. It is a small, fast model for high-volume work such as summaries, classification, database queries, customer support, browser use and acting as a subagent for larger models. Anthropic says it costs around 75% less to run than Haiku 4.5 and scores far higher on most benchmarks. The post also halves Sonnet 5.5 cache-read prices and adds monthly API credits for Max and Team subscribers.

## Key numbers

- $0.10 / $0.50: Haiku 5.5 input and output price per million tokens for prompts up to 100k, vs $1.00 / $5.00 for Haiku 4.5
- ~75%: average reduction in cost to run vs Haiku 4.5, per Anthropic
- 72.4% vs 15.7%: OSWorld 2.1 offline subset, Haiku 5.5 vs Haiku 4.5
- ~20%: cost cut for Sonnet 5.5 on most agentic work after halving cache-read prices

## What it reports

- Pricing per million tokens (prompts up to / over 100k): input $0.10 / $0.50, output $0.50 / $2.50, cache reads $0.01 / $0.05. Haiku 4.5: $1.00 input, $5.00 output.
- Prompts up to 100k tokens made up about 90% of Haiku 4.5 requests; Haiku 5.5 is priced 90% lower there and 50% lower above.
- Benchmarks (Haiku 5.5 / Haiku 4.5 / Sonnet 5.5): GDPval-AA v2.1 1620 / 735 / 1840; OSWorld 2.1 72.4% / 15.7% / 83.9%; Terminal-Bench 4.0 39.2% / 0.0% / 70.6%; Humanity's Last Exam (no tools) 45.9% / 10.2% / 56.9%.
- First Haiku model with an adjustable effort setting.
- Customer figures: AlphaSense 0.84 vs 0.76 on 400 queries (a feature doing about 8M calls a week); HubSpot 92.8% on a CRM suite averaged over three runs; Asana over 30% lower latency; Box 11 points higher than Haiku 4.5 at about half the latency.
- Sonnet 5.5 cache reads drop from $0.20 to $0.10 per million tokens, about 20% cheaper on most agentic tasks.
- Monthly API credits: $100 (Max 5x), $200 (Max 20x), up to $500 pooled (Team).

## How they got it

Benchmarks were run by Anthropic, with methods in the Haiku 5.5 system card; GDPval-AA is from Artificial Analysis. Customer figures come from early-access partners' own evaluation suites.

## Limits

- All numbers come from the vendor or launch partners; no independent replication at publication.
- Anthropic itself says Sonnet 5.5 and Opus 5.5 remain better for complex agentic coding (39.2% vs 70.6% on Terminal-Bench 4.0).

## Why it matters for companies

Most enterprise AI volume is routine work such as summaries, routing, extraction and lookups, and that tier just got about ten times cheaper per token. The design the post describes, a large model planning while a small one does lookups, is now cheap enough to run at scale. Teams should re-price their workloads and test whether a small model is good enough for each step, instead of sending everything to the largest model.

Original: https://www.anthropic.com/claude-haiku-5-5