Anthropic on Wednesday introduced Claude Haiku 5.5, describing it as the company’s cheapest, fastest, and most capable small model so far. The release targets high-volume workloads such as document summarization and database queries, as well as latency-sensitive uses including real-time customer service and browser actions performed on behalf of users. For prompts under 100,000 tokens, Anthropic set pricing at $0.1 per million input tokens and $0.5 per million output tokens, matching the price of OpenAI’s competing small model GPT-6 Luna, which was released on Sept. 22. Compared with Haiku 4.5, whose equivalent rates were $1 and $5, the new model cuts pricing by 90%, while prompts above 100,000 tokens receive a 50% discount. Anthropic said roughly 90% of requests on the older model fell below that threshold, and because Haiku 5.5 uses slightly more tokenization, it estimates average savings at about 75%. In benchmark tests, Haiku 5.5 posted a 72.4% success rate on OSWorld 2.1 versus GPT Luna’s 48.9%, and scored 39.2% on Terminal-Bench 4, above Luna’s 16.4% and Haiku 4.5’s 0%, though below Sonnet 5.5’s 70.6%. It is also the first Haiku model with an adjustable effort setting, allowing users to trade off cost against answer quality.
Anthropic on Wednesday released Claude Haiku 5.5, calling it the company’s cheapest, fastest, and most capable small model to date.
The model is aimed at high-volume tasks such as document summarization and database queries, along with speed-sensitive use cases including real-time customer service and browser actions carried out on behalf of users.
Pricing set at GPT-6 Luna levels
For developers integrating the model into their own products, prompts under 100,000 tokens are priced at $0.1 per million input tokens and $0.5 per million output tokens. Anthropic said that matches the pricing of OpenAI’s competing small model, GPT-6 Luna, released on Sept. 22.
By comparison, Haiku 4.5 was priced at $1 per million input tokens and $5 per million output tokens. On that basis, the new rates are 90% lower. For prompts above 100,000 tokens, pricing is reduced by 50%.
Anthropic estimated average savings of about 75%, saying roughly 90% of requests on the older model were below that threshold and that Haiku 5.5 produces a slightly higher token count.
Benchmark results
In benchmark tests, Haiku 5.5 recorded a 72.4% success rate on OSWorld 2.1, ahead of GPT Luna’s 48.9%. On Terminal-Bench 4, it scored 39.2%, above Luna’s 16.4% and Haiku 4.5’s 0%, but below Sonnet 5.5’s 70.6%.
Adjustable effort setting
Anthropic said this is the first Haiku model to include an adjustable effort setting, giving users a way to balance cost against answer quality.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
