Microsoft introduced new artificial intelligence capabilities for its Copilot assistant on Monday. The company launched Critique, a multimodal deep research system that utilizes two AI models working in tandem. One model is responsible for generating responses, while the other reviews and optimizes the output. Microsoft stated that this architecture surpasses traditional single-model approaches and delivers top-tier deep research quality.

The company also introduced Council, an alternative method that runs models from Anthropic and OpenAI simultaneously to generate separate reports. A judge model is then used to create a condensed summary that highlights points of agreement and disagreement. Both features are now available within Microsoft's Frontier program.

Benchmark tests indicate that Critique provides superior deep research quality compared to products from OpenAI, Google, and Perplexity, according to a social media post by Chief Executive Officer Satya Nadella. Nicole Herskowitz, a company vice president, stated that this multi-model approach should help limit AI hallucinations and enhance productivity.