Gemini API

What it covers

The Gemini API supports text and multimodal model calls, streaming, embeddings, file inputs, tools, and live sessions. Google AI Studio provides project and key management for direct Gemini API use.

Request shape

curl "https://generativelanguage.googleapis.com/v1beta/models/YOUR_MODEL:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Compare these two API responses."}]}]}'

Use the official model list at integration time. Quotas belong to the project, so creating more keys does not multiply the available allowance.

Key endpoints

Endpoint	Description
POST /v1beta/models/{model}:generateContent	Generate text content
POST /v1beta/models/{model}:streamGenerateContent	Streaming generation
POST /v1beta/models/{model}:embedContent	Generate embeddings
POST /v1beta/models/{model}:countTokens	Count tokens
GET /v1beta/models	List available models
GET /v1beta/tunedModels	List tuned models
POST /v1beta/tunedModels	Create tuned model

Decision notes

Separate stable integration facts from model-specific facts. The base API and authentication pattern are relatively stable; model IDs, preview status, token limits, prices, and quotas need dated verification.

Python example: generate content

import httpx

API_KEY = "your_gemini_api_key"
BASE = "https://generativelanguage.googleapis.com/v1beta"

def generate(prompt: str, model: str = "gemini-2.5-flash") -> str:
    """Generate content. API key goes in the query string."""
    resp = httpx.post(
        f"{BASE}/models/{model}:generateContent",
        params={"key": API_KEY},
        json={"contents": [{"parts": [{"text": prompt}]}]},
        timeout=60,
    )
    if resp.status_code == 429:
        raise RuntimeError("Rate limited — back off and retry")
    resp.raise_for_status()
    data = resp.json()
    return data["candidates"][0]["content"]["parts"][0]["text"]

print(generate("What is a dividend? One sentence."))

Notes:

• Auth is the key query parameter (or x-goog-api-key header)
• Model names evolve fast; gemini-2.5-flash is the current efficient default — verify in docs
• Free tier has strict RPM/TPD limits; check candidates[].finishReason for MAX_TOKENS / SAFETY
• Use the official google-genai SDK for structured output and function calling

For manually reconstructed function-call history, see the Gemini thought_signature 400 case. It explains how to preserve parallel calls in their original model response.