Berkeley Function-Calling Leaderboard

The de-facto standard tool-use/function-calling eval (Gorilla team: Patil, Stoica, Gonzalez) — AST-based checking, cited in OpenAI, Google, Anthropic, and Qwen model cards; V3 added multi-turn (2024-09), V4 agentic settings (2025). The Gorilla ecosystem (RAFT retrieval-aware finetuning, GoEx safe-execution runtime) feeds it.