Microsoft is now pitting GPT against Claude inside Copilot to fix AI’s lying problem, but is a self-correcting bot worth the new premium price tag?
Microsoft is fundamentally changing how its AI works by allowing rival models to converse.
In a major update to Copilot released today, the tech giant introduced a feature called “Critique” that forces OpenAI’s GPT and Anthropic’s Claude to collaborate on a single task. It is a striking admission that no single AI model is currently perfect enough to handle the complex demands of enterprise work alone.
The new workflow functions like a high-speed editorial desk.
When a user submits a research query, GPT drafts the initial response while Claude simultaneously reviews it for accuracy and citation quality. This “model council” approach has reportedly led to a double-digit improvement in research quality, pushing Microsoft ahead of standalone tools from Google and Perplexity.
By layering these models, Microsoft aims to resolve the industry’s biggest headache: the tendency for AI to hallucinate facts.
Beyond better research, Microsoft is also pushing Copilot Cowork into early access. It’s a much-needed pivot to autonomous agents.
Earlier versions of Copilot focused on email summaries, and Cowork changed that. It will actually do the work, like reconciling budgets or organizing entire project timelines.
But this intelligence comes with a price tag.
Microsoft is simultaneously pulling the free version of Copilot from core Office apps and reserving the integrated experience for paid commercial subscribers. It’s clearly a strategic step.
The tech giant is no longer interested in just giving AI away for fun. And now it’s actively betting that businesses will pay a premium for a “coworker” that finally knows how to check its own work.
