Anthropic has announced Claude Haiku 5.5, a new AI model which the firm says is its cheapest and most capable small model offering to date.
Claude Haiku 5.5 is approximately 75 per cent cheaper than its predecessor Haiku 4.5, Anthropic said, while offering significant jumps in performance. For prompts up to 100,000 tokens, the new model is 90 per cent cheaper than its predecessor.
It has specifically positioned Claude Haiku as a cheaper, more performant alternative to OpenAI’s cheapest offering, GPT-6 Luna.
In the coding benchmark FrontierCode v1.1, which tests models on their ability to produce high quality code in real-world scenarios, Anthropic said Claude Haiku 5.5 scored 46.4 per cent, ahead of the 42.4 per cent scored by GPT-6 Luna.
For autonomous computer use tasks, as measured by the benchmark OSWorld 2.1, Claude Haiku 5.5 scored 72.4 per cent, compared to 48.9 per cent for GPT-6 Luna and 15.7 per cent for Claude Haiku 4.5.
Claude Haiku is the lightweight alternative to Claude Sonnet and Claude Opus, intended for cost-sensitive, repetitive workloads such as classification, summaries or as a sub-agent focused on a specific task.
The independent AI benchmarking platform Artificial Analysis ranks Claude Haiku roughly on par with Moonshot AI’s Kimi K3 and ahead of both GPT-6 Luna and Google’s Gemini 3.8 Flash.
Anthropic customers with early access to the model praised it for its performance, speed and low cost.
“Our customers use Box AI across large volumes of their enterprise content,” said Yashodha Bhavnani, vice president of AI products at the cloud content management firm Box.
“With widespread usage comes the need to manage efficiency and cost, and to find the best model to suit the task at hand. In early testing, Claude Haiku 5.5 scored 11 points higher than Haiku 4.5 at about half the latency. We’d put it to use on analytical work that runs at scale, from cost reports to financial summaries and weekly recurring reviews.”
Claude Haiku 5.5 is also the first in the Haiku lineup to support effort settings. In internal testing, Anthropic said Claude Haiku 5.5 on low effort offers equivalent performance to GPT-6 Luna on high effort.
Anthropic prices the model at $0.10 per million input tokens and $0.50 per million output tokens. Prompts over 100,000 tokens raise these costs to $0.50 per million input and $2.50 per million output respectively.
Cache hits, meaning AI inference that draws on previously processed context, are substantially cheaper, at $0.01 per million input tokens and $0.125 per million output tokens.
Enterprises in particular are looking for low-cost AI models to use for repetitive and less sensitive tasks such as producing boilerplate code or document summarisation, with Chinese AI models increasingly filling this gap.
