OpenFunctions

Gorilla OpenFunctions is a drop-in open-source alternative to proprietary function-calling APIs, designed for developers who need to reliably map natural-language prompts to structured API calls without vendor lock-in. It targets engineers building agentic workflows, tool-using LLM applications, or multi-provider integrations who want a single, unified interface for function calling. Created by the Gorilla LLM team at UC Berkeley, OpenFunctions returns correctly formatted function calls given a prompt and API definition, supporting parallel function calling—a capability that allows the model to output multiple function invocations in a single response, as noted in community discussions comparing tools versus functions performance. The model scores 59.22% on the Berkeley Function Calling Leaderboard (BFCL), placing it 7th overall alongside GPT-5, according to Klavis AI's 2025 benchmark analysis. This open-weight model is free to use and can be self-hosted or accessed via Hugging Face, making it a practical choice for teams that need predictable, auditable function-calling behavior without per-token costs.

OpenFunctions works by accepting a user prompt and an API schema (e.g., a JSON description of available functions) and outputting a correctly formatted function call—including arguments—that can be directly executed. The model supports parallel function calling, meaning it can generate multiple distinct function invocations in one turn, which is critical for complex agentic tasks like querying multiple data sources simultaneously. It is built on the Gorilla framework and is available as gorilla-llm/gorilla-openfunctions-v2 on Hugging Face, with the Berkeley Function Calling Leaderboard providing evaluation code and rankings. The unified API format means developers write one integration instead of adapting to each provider's quirks, and the built-in model routing and ranking (inspired by OpenRouter's approach) simplifies provider selection. However, the model's BFCL score of 59.22% indicates it trails top proprietary models like GPT-5 (also 59.22%) and others that score higher, reflecting a trade-off between openness and raw accuracy.

In the market, OpenFunctions competes directly with OpenAI's Function Calling API, LangChain's tools/agents, and OpenRouter. OpenAI's solution is faster and easier to use per Reddit users, but it is proprietary and incurs per-token costs. LangChain's tools/agents offer flexibility but require more boilerplate and configuration. OpenRouter provides a single key and unified API format with built-in routing, but it is a gateway service, not a model you can self-host. OpenFunctions differentiates by being fully open-source and free, with a drop-in design that mimics the same API pattern as proprietary solutions, enabling easy migration. Its 7th-place BFCL ranking shows it is competitive but not state-of-the-art; for example, GPT-5 and other closed models achieve higher accuracy on complex function-calling benchmarks, making OpenFunctions better suited for cost-sensitive or privacy-conscious deployments where top-tier accuracy is not mandatory.

The honest trade-offs: OpenFunctions' 59.22% BFCL score means it will make more errors than top proprietary models on nuanced or multi-step function calls, potentially requiring fallback logic or retry mechanisms. Being open-source, it lacks the managed infrastructure, automatic updates, and SLA guarantees of cloud APIs—users must handle hosting, scaling, and model updates themselves. Parallel function calling is supported, but the model's reliability in generating correct parallel invocations may be lower than OpenAI's tools endpoint, which was specifically designed for this. Finally, the ecosystem around OpenFunctions is smaller than LangChain's or OpenAI's, meaning fewer community integrations, tutorials, and third-party tooling, which can increase development time for complex agentic systems.

Free and open-source; no per-token costs or licensing fees. Self-hosting requires compute resources (e.g., GPU inference).