Microsoft's AI Chief calls the Anthropic model extremely expensive, aiming to rely on self-developed models to completely eliminate Token purchase costs

According to CogX Beating Monitor, Microsoft's AI lead Mustafa Suleyman openly stated in an interview with Bloomberg that Anthropic's model is priced extremely high, leading many users to urgently seek alternative solutions. Suleyman revealed that Microsoft is currently required to pay a substantial amount to Anthropic, with the team's explicit goal being to reduce and eventually eliminate the associated procurement costs.

To break free from reliance on external state-of-the-art models, Microsoft unveiled 7 in-house MAI models at the June Build 2026 developer conference. The newly released models include the inference-capable MAI-Thinking-1, with Microsoft claiming the new models can significantly reduce Token costs while ensuring performance, and compete with Anthropic's flagship model Claude 4.6 Opus in tasks such as programming.

The urgency to control costs has prompted Microsoft to internally adjust its toolchain. Microsoft is gradually reducing reliance on third-party tools like Anthropic's programming assistant Claude Code, migrating internal R&D and day-to-day development work to the in-house GitHub Copilot CLI to manage internal Token consumption.

Microsoft's acceleration of in-house model development is set against the backdrop of a restructured relationship with OpenAI. The two parties signed a revised agreement on April 27, 2026, ending Microsoft's exclusive licensing of OpenAI's technology and ceasing the revenue share Microsoft paid to OpenAI. The agreement restructuring paves the way for Microsoft to independently develop cutting-edge models and compete directly with OpenAI, Google DeepMind, and Anthropic.