Anthropic PBC released Claude Haiku 5.5 on Tuesday, a small AI model priced at roughly one quarter of what it costs to run its predecessor, Haiku 4.5. The company also cut the price of cache reads on Claude Sonnet 5.5 in half. Haiku 5.5 becomes the third model in the 5.5 generation, arriving two weeks after Opus 5.5 launched on Sept. 22.
Anthropic is directing Haiku 5.5 at repetitive, high-volume work. Summaries and classification tasks are the primary targets, and coding teams can deploy the model as a subagent that Opus 5.5 or Sonnet 5.5 assigns smaller tasks to. Since no Anthropic model runs faster at standard speed, the company recommends Haiku 5.5 for live customer support and browser automation.
The lower running cost comes from a steep cut to list prices. Haiku 4.5, released last October, costs $1 per million input tokens and $5 per million output. For prompts up to 100,000 tokens, which Anthropic said covered about 90% of the older model's requests, Haiku 5.5 charges 10 cents per million input tokens and 50 cents per million output tokens, a 90% reduction on both. Longer prompts receive a 50% discount. The 75% average saving Anthropic cites factors in a new tokenizer that uses slightly more tokens per task. Haiku 5.5 is also the first Haiku model with an adjustable effort setting, allowing further cost tuning.
The 10-cent and 50-cent rates match what OpenAI Group PBC charges for GPT-6 Luna, its low-cost model launched last month. Anthropic's published benchmarks place Haiku 5.5 ahead of Luna on all six tests where both have scores. On OSWorld 2.1, which measures agents operating a real computer through long multistep tasks, Haiku 5.5 scored 72.4% on the offline subset versus 48.9% for Luna. On the Terminal-Bench 4.0 agentic coding test, Luna scored 16.4%, less than half Haiku 5.5's 39.2%. For complex agentic coding, Anthropic still points customers to Sonnet 5.5 and Opus 5.5.
Asana Inc. tested Haiku 5.5 before release through the evaluation suite for its AI Teammates agent. Task completion latency came in more than 30% lower than with the model Asana uses today, and inference on each agent turn ran up to 2.5 times faster. Aaron Vinh, a staff software engineer at Asana, described the experience as noticeably snappier.
Anthropic said alignment testing found far fewer instances of misaligned behavior than Haiku 4.5 showed. Cybersecurity safeguards allow more defensive work than Sonnet 5.5 permits, though penetration testing and other techniques attackers favor remain blocked. Organizations needing wider access can apply to the Cyber Verification Program Anthropic expanded on Tuesday.
The Sonnet 5.5 price cut affects cache reads, which drop from 20 cents per million tokens to 10 cents. Sonnet 5.5 launched Sept. 28. Because cached tokens make up a large share of what models consume, Anthropic expects the reduction to take about 20% off the cost of most agentic work on that model. Haiku 5.5 is available now on the Claude Platform as claude-haiku-5-5 and through Amazon Web Services Inc., Google Cloud and Microsoft Azure.
More AI news from TechManNews.
