Anthropic has launched Claude Haiku 5.5, a compact model featuring a 1 million token context window. The model is accessible via the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and the Claude Platform on AWS. It supports both text and image inputs, with a knowledge cutoff date of June 2026.
A notable addition is the adjustable effort setting, with adaptive thinking enabled by default. The model uses a new tokenizer that counts text as approximately 30% more tokens compared to the previous Haiku 4.5 version. Developers should note that non-default temperature, top_p, or top_k values currently trigger a 400 error.
Pricing is tiered based on prompt length. For prompts up to 100K tokens, input costs $0.10 and output costs $0.50 per million tokens. For prompts exceeding 100K tokens, rates increase to $0.50 for input and $2.50 for output. Cache reads are priced at $0.01, while 5-minute cache writes cost $0.125.
Anthropic estimates that Haiku 5.5 is roughly 75% cheaper than its predecessor, Haiku 4.5, which charged $1 for input and $5 for output. Batch processing offers an additional 50% discount. While GPT-6 Luna offers competitive pricing for short contexts, Anthropic continues to recommend Sonnet 5.5 and Opus 5.5 for complex agentic coding tasks.
