Product Launches

Microsoft's Foundry IQ adds self-improving AI agents, 50+ model pool

Share
Microsoft's Foundry IQ adds self-improving AI agents, 50+ model pool

Microsoft has unveiled a self-improvement loop for AI agents and expanded its in-house model lineup with the official launch of Decision-1.1. The updates are part of the company's enhanced enterprise AI development platform, Foundry IQ.

TL;DR

  • Microsoft's Foundry IQ now includes a self-improvement loop for AI agents, allowing them to refine performance based on benchmarks and human feedback.
  • The platform supports a model pool of the top 50 models, including new in-house models like Decision-1.1, MAI-Code-2, and MAI-Image-2.7.
  • These updates aim to streamline AI development, improve decision-making, and provide more tailored solutions for enterprises.

What happened

On November 17, Microsoft announced significant updates to its enterprise AI development platform, Foundry IQ. The platform now features a self-improvement loop for AI agents, enabling them to assess and enhance their performance based on quality, safety, response speed, and cost benchmarks. Human experts can review an agent's work logs and provide feedback, which is then integrated into the improvement loop.

Foundry IQ also introduces a feature that allows companies to use their own data to reinforce-train models, refining them for specific operational needs. The platform provides the necessary computing resources for this training.

Microsoft officially launched its decision model, Decision-1.1, designed to make situation-appropriate judgments at a low cost. Unlike large language models that spend more time on tasks like classification and prioritization, Decision-1.1 focuses on producing quick answers within a defined scope and provides a reliability score. The model was developed based on Google Gemma 4 26B A4 and supports image analysis.

Why it matters

For developers and startups, these updates streamline the AI development process, providing tools to build, test, and deploy AI applications more efficiently. The self-improvement loop and human feedback integration can lead to more robust and reliable AI agents.

Investors may see these enhancements as a strategic move to strengthen Microsoft's position in the enterprise AI market. The expanded model lineup and the inclusion of top models from various providers can attract a broader range of users and industries.

However, the effectiveness of the self-improvement loop and the decision model's reliability in real-world scenarios remain to be seen. The competitive landscape, with other major players like OpenAI, Anthropic, and Meta, also poses challenges.

Key facts

  • Foundry IQ now includes a self-improvement loop for AI agents.
  • The platform supports a model pool of the top 50 models.
  • Decision-1.1 is designed for quick, situation-appropriate judgments with a reliability score.
  • MAI-Code-2 and MAI-Code-2-Flash are new in-house models for demanding and routine coding tasks, respectively.
  • MAI-Image-2.7 and MAI-Image-2.7-Flash are new image models available in preview.
  • MAI-Cyber-2-Flash is a security-focused model available in preview in the agent-based code security system codenamed MDASH.
  • Foundry IQ provides more than 11,000 models built by OpenAI, Anthropic, Meta, Mistral AI, and open-source communities.

Context

Microsoft's updates to Foundry IQ and the expansion of its in-house model lineup reflect the ongoing competition and innovation in the AI industry. As enterprises increasingly adopt AI solutions, platforms like Foundry IQ that offer comprehensive tools and a wide range of models are becoming more valuable.

The self-improvement loop and the decision model's focus on quick, reliable judgments address some of the key challenges in AI development, such as ensuring quality, safety, and efficiency. These updates position Microsoft as a strong contender in the enterprise AI market, alongside other major players.

Topics

Related coverage

Join the discussion

Have a take on this story? Weigh in with our community on Facebook.

💬 Discuss on Facebook →