Anthropic launches claude haiku 5.5: a cost-effective model with 1m context window

Anthropic has launched Claude Haiku 5.5, its most affordable and efficient small model to date. The model supports up to 1 million input tokens and 128,000 output tokens, with pricing starting at $0.10 per million input tokens and $0.50 per million output tokens. This represents a 90% reduction in cost compared to Claude Haiku 4.5 for prompts up to 100K tokens. Haiku 5.5 is available via the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and the Claude Platform on AWS. It features an adjustable effort setting, with adaptive thinking enabled by default. The model has a June 2026 knowledge cutoff and supports batch jobs with up to 300K output tokens in beta. Anthropic estimates Haiku 5.5 is about 75% cheaper on average than Haiku 4.5, with batch processing further reducing costs by 50%. Non-default temperature, top_p, or top_k values result in a 400 error, and the new tokenizer counts text as 30% more tokens than Haiku 4.5. Haiku 5.5 is recommended for high-volume tasks such as summaries, compaction, classification, and subagent work. Pricing splits at 100K prompt tokens, with input costs rising to $0.50 and output to $2.50 above that threshold. GPT-6 Luna offers similar short-context rates but becomes more expensive above 272K input tokens. Sonnet 5.5 remains the top-performing model, achieving 70.6% on Terminal-Bench 4.0. Anthropic advises using Sonnet 5.5 and Opus 5.5 for complex agentic coding.