Gorilla

Gorilla is a standout in the niche of LLM-powered function calling, offering a mix of performance and openness that few can match. Its OpenFunctions v2 model is genuinely competitive with GPT-4 on the BFLC leaderboard, which itself is a valuable tool for evaluating other models. We'd reach for Gorilla when building agents that must call many APIs or execute code, especially where safety matters—GoEX's undo and damage confinement are rare among open-source solutions. However, it's not plug-and-play; you'll need to handle deployment and prompt engineering yourself, and the 6.91B parameter size may be less efficient than larger dense models. Compared to alternatives like GPT-4 function calling, Gorilla is free but requires more setup; versus open-source models like CodeLlama, it specializes in API generation. In practice, the CLI tool and Colab notebook lower the barrier, but real production use demands serious infrastructure. Where it bites: limited community support beyond Discord, and the model can sometimes miss nuanced function calls that proprietary models handle with native tools. Still, for a research group or enterprise wanting to avoid vendor lock-in, Gorilla is a strong bet.