During the keynote presentation at Salesforce's Dreamforce conference on September 15, 2026, in San Francisco, Anthropic CEO Dario Amodei made headlines with the launch of a new AI model called Sonnet 5.5, which marks the company's second release following Amodei's recent call to ease the rapid advancement of AI technologies.
Sonnet 5.5 is positioned as a more cost-effective and faster alternative, boasting enhancements in coding capabilities, task completion, and the creation of refined documents, slides, and spreadsheets compared to its predecessor. This unveiling comes just days after the introduction of Opus 5.5, a pricier model, while Anthropic announced that the most budget-friendly model in its lineup, Haiku 5.5, will be released soon.
In a conversation with CNBC, Theo Chu, Research Product Manager at Anthropic, explained, “Sonnet is designed for budget-conscious users who may not require extensive intelligence. It's ideal for routine tasks that need execution without the nuanced judgment that Opus can provide.”
Recent weeks have seen heightened focus on the safety protocols of AI companies, including Anthropic, amidst increasing concerns from researchers about the potential dangers associated with AI advancements. Earlier this month, Amodei surprised many in the technology and governmental sectors by advocating for a slower pace in the evolution of advanced AI models.
According to a blog post from Anthropic, the company clarified that Sonnet 5.5 does not push the boundaries of model capabilities. The majority of its alignment testing has targeted a specific range of risks applicable to models at any performance level. In AI terminology, a well-aligned model acts in harmony with human values and interests.
Nonetheless, Anthropic highlighted that Sonnet 5.5 presents significant enhancements in cybersecurity features compared to its earlier model. This iteration is also the first in the Sonnet series to incorporate advanced fallback protocols and cyber safeguards akin to those found in the company's leading models.
“From the outset, prioritizing alignment and safety has been central to our mission,” remarked Chu. “It’s a focus we've consistently upheld across all our models.”
Sonnet 5.5 is available for $2 per million input tokens and $10 per million output tokens, which is half the cost of Opus 5.5. Notably, the new model also consumes fewer tokens per task than the previous version, allowing for reduced operational costs.
As of now, Sonnet 5.5 can be accessed on all major platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.
