Jakarta, INTI - The competition between OpenAI and Anthropic has entered a new, more collaborative phase through cross-platform integration. Microsoft, along with the developer community, reportedly has begun combining the GPT and Claude models to produce more accurate output than either model would produce separately.
Microsoft is introducing two new features in Copilot Researcher: Critique and Council, which leverage the strengths of both models to reduce misinformation and inaccurate references in complex research. In Critique mode, GPT is responsible for planning, evaluating sources, and creating initial drafts, while Claude acts as a reviewer, verifying accuracy and completeness before final results are delivered.
In contrast, the Council feature runs GPT and Claude simultaneously in a single task. Both models generate reports independently, while a third model acts as an assessor, summarizing the similarities and differences between the two results. Microsoft explains that this approach divides the roles between the author and review models to improve output quality. Currently, this feature is only available to users of the Microsoft Frontier program with a Microsoft 365 Copilot license.
Integration in Software Development
This collaboration also extends to software development through an OpenAI plugin for Claude Code. This plugin allows OpenAI's Codex agent to be integrated directly into Anthropic's developer ecosystem without disrupting workflows. Several key commands are available, such as /codex:review for general review, /codex:adversarial-review for more critical evaluation, and /codex:rescue, which takes over when the system experiences problems.
Technically, this plugin does not run standalone, but instead relies on the locally installed Codex command line. Users need a ChatGPT or OpenAI API subscription, as well as Node.js version 18.18 or later to run it.
This collaborative approach is based on the understanding that each AI model has its own limitations. By combining models from different providers, the evaluation process becomes more independent and produces more optimal output, both for professional research and everyday software development.
Conclusion
OpenAI and Anthropic began collaborating by integrating Microsoft's GPT and Claude models into Copilot features like Critique and Council to improve research accuracy. This approach divides the roles of model creator and reviewer and extends to software development through the Codex plugin for Claude Code. This cross-AI collaboration is considered capable of reducing bias and errors because each model complements the other's weaknesses.
Read more: Bappenas Promotes Digital and AI Investment Cooperation with Huaxia Kunpeng