TL;DR

Anthropic released Claude Haiku 5.5 on Wednesday 7 October, a small model for repetitive, high-volume jobs that it says is roughly 75% cheaper to operate than Haiku 4.5, on average. The launch came with a halved cache-read price for Sonnet 5.5 and new monthly API credits for Max and Team subscribers.

Price and pitch

Haiku 5.5 sits below Sonnet and Opus in the 5.5 line-up. Anthropic aims it at summaries, context compaction, database queries and classification, and suggests running it as a subagent beneath the larger models on coding work. It calls Haiku 5.5 its fastest model so far at standard speed, though Opus in Fast Mode still beats it.

Below the 100,000-token prompt mark, where Anthropic says nine in ten Haiku 4.5 requests fell, the price is $0.10 per million input tokens and $0.50 per million output, a 90% cut. Above that threshold it is $0.50 and $2.50, half the old rate. A new tokeniser needs a few more tokens for the same job, which is why the blended saving comes out nearer 75%.

It is also the first Haiku model whose effort level users can adjust, trading cost against capability.

Benchmarks

On Anthropic’s own figures the jump over Haiku 4.5 is large: 72.4% on the OSWorld 2.1 computer-use test against 15.7%, and 39.2% on Terminal-Bench 4.0 where the older model scored zero. Sonnet 5.5 still leads at 70.6% on Terminal-Bench, and Anthropic says its bigger models remain the better choice for complex agentic coding.

The table also includes OpenAI’s GPT-6 Luna, the model OpenAI is now putting in front of free ChatGPT users. Haiku 5.5 beats it on every test where both were scored, though these are vendor-run results rather than independent ones.

The other changes

Sonnet 5.5 cache reads drop from $0.20 to $0.10 per million tokens, which Anthropic says makes most agentic tasks about 20% cheaper. Starting this week, Max 5x subscribers get $100 a month in API credit, Max 20x subscribers $200, and Team plans up to $500 shared across users. Anthropic’s Python and TypeScript SDKs gain beta support for computer and browser use.

Looking forward

Haiku 5.5 is available through Microsoft Azure, AWS and Google Cloud as well as Anthropic’s own platform. For UK teams running large volumes of routine classification or summarisation, the per-token cut is big enough to justify re-testing jobs that were previously too costly to automate, as long as they are checked on real tasks rather than benchmark tables.