For AI to go beyond chatting in order to actually get things done, it needs to know how to use real tools like apps or services on the internet. This process is called tool-calling, and it’s one of the most important skills for an AI agent. But teaching AI how to do this isn’t easy, as it takes a lot of good, real-life examples, and those are hard to find. That’s why tech giant IBM (IBM) and the University of Washington created Toucan. Interestingly, Toucan is now the largest public collection of tool-using examples for AI, with 1.5 million real-world tasks involving over 2,000 different online tools.
Elevate Your Investing Strategy:
- Take advantage of TipRanks Premium at 50% off! Unlock powerful investing tools, advanced data, and expert analyst insights to help you invest with confidence.
These examples include practical scenarios like analyzing reports, sending calendar invites, or drafting summaries, and were built using real metadata from API servers found on GitHub and Smithery.ai. Notably, the researchers filtered out tools that didn’t work and used five LLMs to create task plans involving one or more tools. In addition, other models were used to simulate how an AI agent would complete each task step-by-step, while a separate group of models rated each scenario for difficulty and quality, which helped the team choose the best examples.
Moreover, initial testing shows that Toucan-trained models performed strongly on well-known benchmarks. For instance, a Qwen-2.5 model (32B) fine-tuned on Toucan improved by nearly nine percentage points on the Berkeley Function Calling Leaderboard (BFCLv3) and slightly outperformed GPT‑4.5-Preview. The model also did well on MCP-Universe, which is a benchmark that covers real-world tasks like financial analysis and web search. As a result, researchers now plan to expand Toucan with newer tools and build a training environment to develop AI agents for enterprise use cases.
Is IBM a Buy, Sell, or Hold?
Turning to Wall Street, analysts have a Moderate Buy consensus rating on IBM stock based on six Buys, six Holds, and one Sell assigned in the past three months, as indicated by the graphic below. Furthermore, the average IBM price target of $287.50 per share implies that shares are trading near fair value.
