Anthropic is getting ready to release a new AI model that combines traditional language model capabilities with advanced reasoning functions.
According to The Information, the model will be available in the coming weeks, with a focus on enterprise applications.
The timing aligns with OpenAI's recent announcement that GPT-5 will arrive in the coming months as a "system" that incorporates reasoning capabilities from their o-models into standard language model functions. OpenAI has decided not to release its o3 reasoning model separately.
Both companies are adjusting their strategies to these hybrid models as traditional language model scaling - adding more data and parameters - shows diminishing returns.
OpenAI hasn't introduced a significantly better language model since GPT-4 in March 2023, and Anthropic has indefinitely delayed its Opus 3.5 flagship model.
While reasoning models can perform significantly better on tasks with clear right or wrong answers, such as coding and mathematics, it is not yet clear if and how they can improve the overall performance of language models, especially in areas such as writing and creation, and gaining new knowledge outside the training data distribution.
Resource control meets enterprise needs
A key feature of Anthropic's new model is its variable resource allocation - users can adjust how much computing power the model uses for each task through a simple slider. At its lowest setting, the model functions as a standard language model without thought chain generation. OpenAI currently limits users to three preset levels for its reasoning models.
According to The Information's sources, early tests suggest that the model performs well in practical programming tasks. One user reports that it handles complex code bases with thousands of files more effectively than OpenAI's o3-mini model, and generates working code more reliably on the first try. OpenAI's o-models maintain an edge in academic programming tasks.
Anthropic forecasts strong growth of its API business
According to The Information, Anthropic projects revenue of up to $34.5 billion by 2027 in an optimistic scenario; their base case projection is $12 billion, which is still a significant upside from its current projection of $3.7 billion for 2025.
The company specifically wants to dominate the API market, projecting revenue three times higher than OpenAI's by 2027 - the same year Anthropic expects to turn its first profit.
Anthropic has already gained ground in AI-assisted software development. An investor analysis shows their market share doubled from 12 to 24 percent, largely thanks to their Sonnet "3.6" model, while OpenAI's share dropped from 50 to 34 percent.
