Claude Haiku 5.5 is out: what changed, and what it costs you

Anthropic calls Haiku 5.5 the cheapest, fastest and most capable small model it has released, built for high-volume work such as summaries, classification, subagents and browser use.^[1]

A prompt of over 100,000 tokens costs more: $0.50 per million input tokens and $2.50 per million output. Anthropic says about 90% of requests to Haiku 4.5 were shorter than that.^[1][4]

Cache reads cost $0.01 per million tokens and cache writes $0.125 per million on prompts up to 100,000 tokens, a tenth of Haiku 4.5's.^[1]

Anthropic says it costs about 75% less to run than Haiku 4.5 on average, counting its newer tokenizer, which reads the same text as about 30% more tokens.^[1][2]

Its context window is 1 million tokens, against Haiku 4.5's 200,000; it can write up to 128,000 tokens in one reply, and its training data runs to June 2026.^[3][6]

It's the first Haiku with an effort setting: it decides when and how much to think, and thinks less at a lower effort.^[1][2]

On Terminal-Bench 4.0, an agentic coding test, Anthropic reports 39.2% for Haiku 5.5, against 0.0% for Haiku 4.5 and 70.6% for Sonnet 5.5, and says Sonnet 5.5 and Opus 5.5 remain the better choices for complex coding.^[1]

Its safety classifiers can decline a request, and unlike Sonnet 5.5 and Opus 5.5 it has no fallback to another Claude model when they do.^[5]

The same day, Anthropic halved Claude Sonnet 5.5's cache-read price, to $0.10 per million tokens.^[1]

It's on the Claude Platform as claude-haiku-5-5, and through Amazon Web Services, Google Cloud and Microsoft Azure.^[1]