Claude Haiku 5.5 — fastest small model, priced to match GPT-6 Luna

Anthropic's fastest small model, priced to match OpenAI's cheapest tier, for summarising, classifying and database work.

Laptop screen showing colourful programming code
Illustrative stock photo — Photo by Mohammad Rahmani on Unsplash. Not a product screenshot and not the lab's own image.

In this article: In brief · What announced · Cost and access · Why it matters · What to check · What is still unknown · At a glance · Sources

LAUNCHED7 Oct 2026
→
PRICE (AS PUBLISHED)$0.10 input / $0.50 output per 1M tokens for requests under 100,000 tokens
ACCESSGeneral

In brief

What Anthropic announced

On 7 Oct 2026, Anthropic released Claude Haiku 5.5 — fastest small model, priced to match GPT-6 Luna for chat / coding work. Reported 1,620 on GDPval-AA knowledge-work benchmark vs 735 for Haiku 4.5 (reported figure, not our test). Positioned as a supporting worker for Opus and Sonnet.

Anthropic’s official announcement calls Haiku 5.5 its cheapest, fastest and most capable small model, built for high-volume, cost-sensitive work such as summaries, classifications, database queries and subagent work for Opus and Sonnet. For prompts up to 100,000 tokens the published price is $0.10 per million input tokens and $0.50 per million output tokens, with $0.50 and $2.50 above that threshold. Anthropic also reports a 1 million-token context window and adjustable effort controls — figures from the maker, not an independent test by this site.

What it costs, and who can use it today

Price: $0.10 input / $0.50 output per 1M tokens for requests under 100,000 tokens. Access: General. Availability: Released 7 Oct 2026, on AWS, Google Cloud and Azure. A token, where prices are per million tokens, is roughly part of a word; prices are per million tokens processed, not per message. Open-weight means downloadable weights — check the licence for commercial use, because downloadable does not automatically mean unrestricted.

Why it matters

This is a price war data point, not just a model. $0.10/$0.50 is now the floor both Anthropic and OpenAI are matching.

What to check before you rely on it

Benchmark figures in this article, where given, are reported figures from the maker or from named coverage — they are not tests run by AI Updates.

What is still unknown

At a glance

LAUNCHED7 Oct 2026
Lab and typeAnthropic · Chat / Coding
Price$0.10 input / $0.50 output per 1M tokens for requests under 100,000 tokens
AccessGeneral
Last checked25 Oct 2026

Frequently asked questions

When was Claude Haiku 5.5 launched? 7 Oct 2026, by Anthropic.

How much does Claude Haiku 5.5 cost? $0.10 input / $0.50 output per 1M tokens for requests under 100,000 tokens

Who can use Claude Haiku 5.5 today? Released 7 Oct 2026, on AWS, Google Cloud and Azure

Sources

Also checked: VentureBeat category page, 9 Oct 2026, plus release calendar — a calendar or category page verifies date, lab and type only, not price or access.

Official sources are listed first, because a lab’s own announcement or product page is the primary source for price, access and availability. Secondary coverage is listed second, and is never used to upgrade an unverified figure into a settled fact.

Corrections: see Corrections for dated corrections already applied to this site, including the Kimi, Muse Wonder, Mistral and Duet corrections. If a price, date or access rule on this page has changed, it will be corrected here with a dated note. Last checked 25 Oct 2026. Prices and availability can change after the checked date — confirm on the provider page.

Related launches

Humanoid robot with a tablet screen

Kimi K3.1 Pro — Moonshot's open-weight flagship, service first, weights to follow

The most recent flagship-class release tracked in October: general, coding and multimodal work from Moonshot AI.

Access: General · Price: Not published in the sources checked

GeneralChat / CodingRead ↗
Hands typing quickly on a laptop keyboard

MiniMax M3.1 Flash — fast tier, price not yet published

A speed-focused release: general use and coding, positioned as the fast tier in MiniMax's line.

Access: Not published · Price: Not published in the sources checked

Not publishedChat / CodingRead ↗
Three people working together around a laptop

Muse Spark Open — Meta's open-weight general model, licence to confirm

Meta's open release on 15 Oct, for general, coding and multimodal use.

Access: Open-weight · Price: Not published in the sources checked

Open-weightChat / CodingRead ↗
← Back to latest