Anthropic reveals Haiku 5.5 model as AI pricing war intensifies

Anthropic ( ANTH.PVT ) took the wraps off its Claude Haiku 5.5 model on Wednesday, the latest in its Claude 5.5 family, promising improved and lower costs than its predecessor.

According to the AI startup, Haiku 5.5 is the company's best lightweight model yet and beats out Haiku 4.5 in coding, computer use, and knowledge work.

Anthropic's Haiku is designed for quick, low-cost use cases, whereas its Sonnet and Opus offer higher-level coding, enterprise, and other capabilities. The company's Fable is its most powerful model and is meant for days-long workloads.

Then there's Anthropic's Mythos , a version of Fable with fewer safeguards, strong cybersecurity, and life sciences characteristics. Mythos is only available to certain members of Anthropic's cyber verification pro gram .

Pricing has become an increasingly important part of the AI conversation; enterprise customers look for ways to reduce costs following the unfettered spending during the brief "tokenmaxxing" era .

Companies are also looking toward open-weight models from US-based and Chinese AI developers to cut down on their AI spending.

Anthropic said Haiku 5.5 costs an average of 75% less than Haiku 4.5. Customers will pay $0.10 per million input tokens and $0.50 per output token. That's in line with rival OpenAI's ( OPAI.PVT ) smallest model, GPT-6 Luna.

Tokens are units of measurement for AI usage and represent parts of or entire words.

Anthropic is on the march toward its highly anticipated initial public offering, expected sometime before Thanksgiving.

Reuters has obtained Anthropic's S-1 investor prospectus, which outlines the company's revenue, losses, risks, and opportunities , and reports that the startup saw sales of $4.6 billion in 2025, up 12x year-over-year. But it lost a whopping $42 billion in the same period.

OpenAI is similarly preparing its public market debut, but has reportedly delayed the action until early 2027.

Both companies are also contending with the fallout from their AI models going rogue and hacking into third-party companies' networks.

More recently, Anthropic CEO Dario Amodei joined other AI CEOs at a White House summit to discuss AI industry self-regulation with President Trump .

Email Daniel Howley at dhowley@yahoofinance.com. Follow him on Twitter at @DanielHowley .