This is one of those places where some vendors are having to finally pass the actual costs of AI inference along to customers, or are at least unwilling to aggressively subsidize them as much as they have been doing. To ever be profitable, they either need a revolutionary shift in the costs of AI inference (massive efficiency increases without capability hits) or they have to get customers so invested in the platforms that they have little choice but to start paying the real costs. Broadly speaking, I don't think Microsoft has as much of a long-term incentive as Google, Anthropic, or OpenAI to keep playing pretend, and that's why they seem to make some of these product choices that seem at odds with others in the market. Claude Cowork is great, but eventually Anthropic is going to be unwilling or unable to keep providing hundreds per month in inference for $20. (Also - that's consumer pricing. If you are doing Team/Enterprise pricing, you are paying for Claude/Cowork at the API usage rates, just like anyone wanting to use M365 Cowork).