Google Gemini

Official positioning

The Gemini API provides native multimodal models that combine text, image, audio, and video, plus embedding, media generation, and live interaction. The official model page defines input and output modalities, context, supported methods, and lifecycle per model; a family name is not a substitute for checking the exact model.

Prefer Interactions for new direct projects

Google made Interactions API generally available and recommends it for all new projects. GenerateContent remains supported but is now Legacy. Interactions places model and input in the body and returns status, usage, and typed steps[], so a contents/parts integration cannot migrate by changing only the URL.

Direct versus gateway authentication and routes

Direct Google uses a Google-issued API key in x-goog-api-key, and new projects can call /v1beta/interactions. This site has no Interactions route; it currently exposes /v1beta/models/{model}:generateContent and :streamGenerateContent. The site recommends Bearer and also accepts a site-issued key in x-goog-api-key on those native routes. Never send a Google key to this gateway.

Responses, streaming, and tools

This site's GenerateContent route uses contents[].parts and returns candidates, finishReason, promptFeedback, and safety information; its streaming route emits incremental candidates. Direct Interactions returns typed steps and different SSE events. Function calling, structured output, files, caching, and safety settings must be verified per model and protocol.

Workloads and production checks

Gemini fits long-context understanding, image/audio/video analysis, coding and reasoning, structured extraction, and low-latency multimodal interaction. Distinguish stable, preview, experimental, and latest identifiers, then verify geography, quota, files, output modality, grounding, tools, data policy, and deprecation notices before production.