Anthropic: Claude Haiku 5.5 (batch)

Claude Haiku 5.5 (batch) is the asynchronous Message Batches API offering of Anthropic’s Claude Haiku 5.5 model (API ID claude-haiku-5-5). It delivers the same capabilities as the synchronous endpoint—adaptive thinking with configurable effort (low through max), a 1M-token context window, up to 128K output tokens (and up to 300K output tokens on the Batch API with the output-300k beta header), and text-and-image input with text output—while processing requests in bulk with a 50% discount on input and output pricing versus standard API calls.

Haiku 5.5 is positioned as the fastest and most cost-efficient model in the Claude 5.5 family, aimed at high-volume, latency-tolerant workloads such as summarization, classification, routing, context compaction, customer support, and subagent coding tasks alongside larger Sonnet or Opus models. Anthropic reports large gains over Haiku 4.5 on agentic coding (for example SWE-Bench Pro and SWE-bench Multilingual), knowledge work (GDPval-AA), computer use, and expert reasoning (Humanity’s Last Exam with tools), with evaluations typically run at max adaptive thinking effort unless noted otherwise.

The batch variant suits offline pipelines, large backfills, and evaluation runs where immediate streaming responses are not required. Knowledge cutoff is June 2026. The model is closed-source and available through the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry; parameter count and internal architecture are not disclosed publicly.