Anthropic released Claude Haiku 5.5 (claude-haiku-5-5) on October 7, 2026, two weeks after Opus 5.5. It is available on the Claude Platform and through AWS, Google Cloud and Microsoft Azure, and Claude Code 2.1.293 made it the default Haiku model the same day.
What changed
- Price. $0.10 input / $0.50 output per million tokens for prompts up to 100,000 tokens; $0.50 / $2.50 above that. Haiku 4.5 was $1 / $5. Anthropic says the cut is 90% for shorter requests and 50% for longer ones, and that 90% of Haiku 4.5 requests were in the shorter tier. Because a new tokenizer "uses slightly more tokens per task," it puts the average saving at around 75%. SiliconANGLE notes the headline rates match OpenAI's GPT-6 Luna.
- Effort setting. The "first Haiku-class model to come with an adjustable effort setting," so developers can trade cost against quality per call.
- Scores (per Anthropic). OSWorld 2.1 offline subset 72.4% (Haiku 4.5: 15.7%; GPT-6 Luna: 48.9%); Terminal-Bench 4.0 39.2% (Luna: 16.4%; Sonnet 5.5: 70.6%). Anthropic still points complex agentic coding at Sonnet 5.5 and Opus 5.5.
- Cheaper Sonnet caching. Sonnet 5.5 cache reads drop from $0.20 to $0.10 per million tokens, which Anthropic expects to cut most agentic workloads on Sonnet by roughly 20%, per SiliconANGLE.
How teams say they are using it
The customer quotes on Anthropic's launch page describe concrete workloads (all customer-reported, not independently verified):
- Asana ran it through the eval suite for AI Teammates, its agent product -- triaging bugs, setting up projects, finding overdue work -- and saw "over a 30% reduction in latency for task completions and up to 2.5x faster inference per agent turn."
- AlphaSense says its Ask in Document feature is "one of our big sources of spend, doing about 8M calls a week in production"; on 400 test queries Haiku 5.5 scored 0.84 against Haiku 4.5's 0.76.
- Box reports it "scored 11 points higher than Haiku 4.5 at about half the latency," and would use it for cost reports, financial summaries and recurring reviews.
- Rogo names "quick lookups, subagents, and summaries"; Cognition adds it as a "sidekick" model in Devin.
Enterprise pattern (analysis): the common shape is a tiered agent -- a large model plans and a small one handles high-volume, narrow steps such as retrieval answers, classification and summarisation. At AlphaSense's volume, a 90% input-price cut changes which features are worth running at all. The constraint is evaluation: every customer quoted ran its own suite before switching, and the tokenizer change means per-task cost, not per-token price, is the number to measure.
Safety note. Anthropic says Haiku 5.5's cyber safeguards are stricter than Haiku 4.5's but allow more defensive work than Sonnet 5.5's; penetration testing stays blocked outside its Cyber Verification Program.
Haiku 5.5 cuts small-model prices by up to 90% and, per Anthropic, closes much of the gap to mid-size models on agent benchmarks; customers such as Asana, AlphaSense and Box describe it as the high-volume tier under bigger planners, but each validated it on its own evals first.