Anthropic's new Claude Haiku 5.5 costs up to 90% less than the last one

Anthropic released Claude Haiku 5.5 on Wednesday, Oct. 7, 2026, the third model in its Claude 5.5 family in about a month. It is the company's cheapest and fastest small model yet, aimed at jobs like summaries, sorting and classifying text, live customer support and acting as a helper agent for bigger Claude models. For most requests it is priced 90% below Haiku 4.5, and Anthropic says it costs about 75% less to run on average.

The headline model gets the press, but the tiny cheap one is what quietly runs inside everything: the support chat, the email summary, the little agent that fetches one number out of a 10-K while a bigger model does the thinking. Cutting that price by up to 90% makes a lot of AI features that were too expensive to run at scale suddenly cheap enough to just switch on. It is also a direct shot in the price war: Anthropic put OpenAI's GPT-6 Luna right in its own benchmark chart, and the timing lands the same day OpenAI pushed GPT-6 to everyone in ChatGPT. Throwing in free monthly API credit for paying subscribers is a nice touch, and also a very clever way to turn hobbyists into customers. The caveat is the usual one: these are Anthropic's own benchmark numbers, and "cheaper per token" only stays cheap if developers do not just run ten times as many agents. Which, let's be honest, they will.

On Wednesday, 7 October 2026, Anthropic released Claude Haiku 5.5 and called it “the cheapest, fastest, and most capable small model we’ve ever released.” It is the third model in the Claude 5.5 family in about a month, after Opus 5.5 and Sonnet 5.5. Reuters notes the lineup expansion comes ahead of Anthropic’s planned IPO. An IPO is a first sale of shares to the public. Reuters does not print a date or a price for that sale. Those lines are Anthropic’s and Reuters’.

What it is for. Anthropic says Haiku 5.5 is built for high-volume, cost-sensitive work: summaries, classification, database queries, live customer support, voice agents, browser use, and serving as a fast helper — a subagent — that bigger models like Opus 5.5 and Sonnet 5.5 hand smaller tasks to. A subagent is a smaller model that does one narrow job while a larger model does the thinking. Classification means sorting text into buckets. Those lines are Anthropic’s.

The price, on the Claude Platform. For prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Longer prompts cost $0.50 and $2.50. A token is a small chunk of text the model reads or writes. Input is what you send. Output is what the model writes back. Haiku 4.5 was $1 and $5. Anthropic says that is 90% lower for prompts up to 100,000 tokens — about 90% of Haiku requests — and 50% lower for longer ones, or about 75% cheaper to run on average. Those figures are Anthropic’s.

On Anthropic’s published benchmarks, Haiku 5.5 scored 72.4% on the OSWorld 2.1 computer-use test, offline subset, versus 15.7% for Haiku 4.5 and 48.9% for OpenAI’s GPT-6 Luna. Sonnet 5.5 scored 83.9% on the same test. OSWorld measures how well a model can operate a real computer to finish long, multi-step tasks. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choice for complex coding work. These scores are Anthropic’s. They are not a rerun by an outside lab.

It is the first Haiku model with an adjustable effort setting, so developers can trade cost against intelligence per task. Effort, here, means how hard the model thinks — and how much you pay — on a given request. That line is Anthropic’s.

Anthropic is also halving the price of cache reads on Sonnet 5.5, from $0.20 to $0.10 per million tokens. A cache read is a saved piece of a long prompt you do not pay full price for again. Anthropic says that makes Sonnet 5.5 about 20% cheaper on most agent work. This week it is adding monthly API credit for subscribers: $100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team plans. Those credits can be used on any Claude model on the Claude Platform. Those lines are Anthropic’s.

Safety, still the company’s. Anthropic says Haiku 5.5 shows major alignment improvements over Haiku 4.5. Alignment means the model stays in line with what people intend. It also says its cybersecurity safeguards still block penetration testing and other attacker-style techniques. Those lines are Anthropic’s.

Where it is on. Anthropic says Haiku 5.5 is available now on all platforms, including the Claude apps for Free, Pro, Max, Team, and Enterprise users, Claude Code, the Claude API under the name claude-haiku-5-5, Amazon Bedrock, Google Cloud, and Microsoft Azure. Amazon confirmed the same-day Bedrock launch. Those lines are Anthropic’s and Amazon’s.

The picture is Anthropic’s official Claude Haiku 5.5 release graphic: the words “Claude Haiku 5.5” across three colored panels. It is the company’s launch image. The frame does not print a calendar date.

In plain terms, Anthropic released its cheapest, fastest small Claude on Wednesday and cut the price by up to 90% versus Haiku 4.5. The model is for high-volume grunt work and helper agents, not the hardest coding jobs. Sonnet 5.5 cache reads got cheaper the same day, and Max and Team subscribers get monthly API credit. The scores and the savings are Anthropic’s.