Gemini 3 API: Built-in Tools + Custom Functions in One Call
Gemini API changelog (March 18): combine built-in tools (Search, Maps, etc.) with custom function calling in a single generateContent request; Grounding with Google Maps supported for Gemini 3. Less glue for multi-tool agents — still Google’s stack and Maps as the geo ground truth.
Agent builders used to orchestrate Google’s built-ins and their own functions in separate passes. The proprietary unlock: one generateContent call with both — plus Maps grounding so geo answers cite Google Maps instead of inventing streets.
Gemini API release notes (March 18) shipped Built-in Tools and Function Calling Combination: specify built-in tools (e.g. Google Search, Maps, file search, code execution) alongside custom functions in a single request. Notes also cite improved context circulation across multi-step built-in tool use. Grounding with Google Maps now supported for Gemini 3 models — location, directions, places, geospatial responses grounded against Maps data with citations. Supported models listed in Gemini API references.
Claims vs checks
Feature availability is Google developer docs primary (changelog). Grounding quality is vendor-claimed reduction in fabricated location detail — not an independent geo-hallucination audit here. Developer Threads/discussions corroborate March 18 timing.
Limits
- Combination still lives inside Gemini API quotas and tool policies.
- Maps grounding is Google-centric ground truth by design.
- Multi-tool reliability depends on model version and prompt structure.
Sources
- Gemini API Release notes, March 18, 2026 — built-in tools + function calling combination; Maps grounding for Gemini 3.
- Developer discussions noting the March 18 rollout.