Thought Leadership
Anthropic has updated its smallest model. Haiku 5.5 offers an adjustable effort setting for the first time, while prompt cache prices for Sonnet 5.5 are also falling.
Anthropic has released Claude Haiku 5.5. The model succeeds Haiku 4.5 and is the third model of Generation 5.5, following Opus 5.5 and Sonnet 5.5. Haiku is the company’s smallest and most affordable model line.
“Claude Haiku 5.5 is designed for large-scale, cost-sensitive tasks.”
Anthropic
As use cases, the company cites summarization, database queries, classification, real-time customer support, browser control, and serving as a sub-agent for Opus 5.5 and Sonnet 5.5 during coding.
Prices Reduced by 90 Percent
For requests up to 100,000 tokens, Haiku 5.5 costs $0.10 for input and $0.50 for output per million tokens. For Haiku 4.5, it was $1 and $5. For longer requests, prices stand at $0.50 and $2.50. According to Anthropic, around 90 percent of requests for Haiku 4.5 fell into the lower-cost category. Since the model uses a new tokenizer and consumes slightly more tokens per task, Anthropic puts the average savings at around 75 percent.
Benchmarks According to Manufacturer Specifications
In the OSWorld 2.1 benchmark, which measures computer control across multi-step tasks, Haiku 5.5 achieves 72.4 percent according to Anthropic, while Haiku 4.5 reached 15.7 percent and OpenAI’s GPT-6 Luna reached 48.9 percent. On Terminal-Bench 4.0 for agentic coding, Haiku 5.5 stands at 39.2 percent, and Sonnet 5.5 at 70.6 percent. For complex programming tasks, Anthropic continues to recommend the larger models. Haiku 5.5 is the first Haiku model with an adjustable effort setting, allowing users to choose between lower costs and higher performance. Independent benchmarks are still pending.
According to Anthropic, the cybersecurity safeguards are stricter than in Haiku 4.5, but less strict than in Sonnet 5.5. Defensive tasks are permitted on a larger scale, while penetration testing is blocked.
Cheaper Cache Read Access for Sonnet 5.5
Additionally, Anthropic is reducing the price of cache read access for Sonnet 5.5 from $0.20 to $0.10 per million tokens. This is intended to lower costs for most agentic tasks by around 20 percent. Subscribers to the Max and Team plans receive a monthly credit for the Claude Platform: $100 for Max 5x, $200 for Max 20x, and up to $500 split across all users for Team. In beta, the SDKs for Python and TypeScript now also support computer and browser control.
Haiku 5.5 is available via the Claude Platform under the model identifier claude-haiku-5-5, as well as through Amazon Web Services, Google Cloud, and Microsoft Azure.
(Editorial Team)
