Claude Haiku 5.5

• Anthropic's cheapest, fastest and most capable small model yet, with the lowest latency in the current Claude lineup
• Much lower list prices: for prompts up to 100K tokens, $0.10 input, $0.50 output and $0.01 cache reads per million tokens, 90% below Haiku 4.5; over 100K tokens, $0.50 / $2.50, still 50% lower; the Batch API saves another 50%
• 1M-token context (Haiku 4.5: 200K) and up to 128K output tokens per request (was 64K); image input; reliable knowledge cutoff of June 2026
• The first Haiku with effort control: low, medium, high, xhigh and max, default medium; adaptive thinking is on by default, and simple requests can be answered without thinking
• A big jump in agentic skills: OSWorld 2.1 computer use from 15.7% to 72.4% and Terminal-Bench 4.0 command-line tasks from 0.0% to 39.2%, both ahead of GPT-6 Luna (48.9% and 16.4%)
• Full tool support: tool use (including forcing a specific tool), structured outputs, prompt caching, the Batch API, the computer use toolset and the new browser use tool
• A pairing Anthropic recommends: inside coding agents led by Opus 5.5 or Sonnet 5.5, Haiku 5.5 runs as a subagent on the many small jobs such as searching, reading and summarizing, cutting overall cost and latency
• Early customer results on Anthropic's launch page: Box measured scores 11 points higher than Haiku 4.5 at about half the latency; Asana cut task-completion latency by more than 30%

Specs and prices are from Anthropic's documentation; benchmark scores are from Anthropic's launch announcement of October 7, 2026. Prices in this table are Anthropic's list prices, not ours. Haiku 5.5 uses a newer tokenizer that counts the same text as roughly 30% more tokens; Anthropic says it still costs around 75% less to run than Haiku 4.5 on average.