Anthropic reports Haiku 5.5 availability and new benchmarks

Anthropic says Haiku 5.5 is available now across platforms including AWS, Google Cloud and Microsoft Azure, and lists `claude-haiku-5-5` as its Claude API model ID.[2][1] GitHub has also made the model generally available in GitHub Copilot.[3] GitHub says early testing found Haiku 5.5 matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps.[3] AWS announced availability on Amazon Bedrock and Claude Platform on AWS.[4] AWS says Bedrock keeps the model’s data within AWS infrastructure and supports regional data residency.[4]

Anthropic says Haiku 5.5 costs around 75% less to run on average, and is priced 90% lower for requests up to 100,000 tokens and 50% lower for longer requests.[1] API input pricing starts at $0.10 per million tokens.[2] However, Haiku 5.5 uses a newer tokenizer: the same text counts as approximately 30% more tokens than with Haiku 4.5.[2] Teams comparing costs should account for that token-count difference alongside the listed price reductions.[1][2]

On Anthropic’s OSWorld 2.1 offline subset, Haiku 5.5 scored 72.4%, compared with 15.7% for Haiku 4.5.[1] Anthropic also reported a score of 1,620 on GDPval-AA v2.1, versus 735 for Haiku 4.5 and 1,437 for GPT-6 Luna.[1] On Terminal-Bench 4.0, Anthropic reported 39.2% for Haiku 5.5, compared with 0.0% for Haiku 4.5 and 16.4% for GPT-6 Luna.[1] These are reported evaluation results, not a guarantee of performance on a team’s own workloads.[1]

Anthropic describes Haiku 5.5 as its fastest model at each model’s standard speed, but says it is slower than Opus models in Fast Mode.[1] The company says the model handles repetitive work such as summaries, compactions, database queries and classification requests.[1] Anthropic also says Haiku 5.5 showed major improvements over Haiku 4.5 across almost all of its alignment evaluations.[1] Its cybersecurity safeguards allow a wider range of defensive work than Sonnet 5.5’s safeguards, but still block penetration testing.[1]