Google Gemini API Not Working? Fix Authentication, Quota & Model Errors

403 API key not valid / permission denied

The most common cause of 403 errors is an API key created in the wrong place or missing the required API enabled. Use this setup:

import google.generativeai as genai
import os

genai.configure(api_key=os.environ["GOOGLE_API_KEY"])
model = genai.GenerativeModel("gemini-1.5-pro")
response = model.generate_content("Hello")

• Create the key at aistudio.google.com — keys created in the Google Cloud Console do not work with the Generative Language API by default.
• Enable "Generative Language API" in the Google Cloud project linked to your key.
• Check for IP or referrer restrictions — if you restricted the key to specific IPs, requests from other IPs will be denied.
• Vertex AI uses different auth — if you are using vertexai, you need Application Default Credentials (ADC), not an API key.

429 Too Many Requests / quota exceeded

Free tier limits: 15 RPM, 1,500 RPD, 1M tokens/day. Add exponential backoff to handle transient quota errors:

import time
import random

def call_with_retry(model, prompt, max_retries=5):
    for attempt in range(max_retries):
        try:
            return model.generate_content(prompt)
        except Exception as e:
            if "429" in str(e) and attempt < max_retries - 1:
                wait = (2 ** attempt) + random.random()
                time.sleep(wait)
            else:
                raise

• Switch to gemini-1.5-flash — the Flash model has a higher free quota than gemini-1.5-pro.
• Cache repeated prompts — if you call the API with identical prompts, cache the response locally.
• Enable billing — upgrading to pay-as-you-go removes most free tier rate limits.

SAFETY finish_reason — blocked response

from google.generativeai.types import HarmCategory, HarmBlockThreshold

response = model.generate_content(
    prompt,
    safety_settings={
        HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_ONLY_HIGH,
    }
)
print(response.candidates[0].finish_reason)  # STOP = success, SAFETY = blocked

• Adjustable categories: HARM_CATEGORY_HARASSMENT, HARM_CATEGORY_HATE_SPEECH, HARM_CATEGORY_SEXUALLY_EXPLICIT, HARM_CATEGORY_DANGEROUS_CONTENT.
• Available thresholds: BLOCK_LOW_AND_ABOVE, BLOCK_MEDIUM_AND_ABOVE, BLOCK_ONLY_HIGH, BLOCK_NONE.
• Free tier restriction: some categories cannot be fully disabled (set to BLOCK_NONE) on the free tier — enable billing first.

Model not found / model name errors

Model IDs must be exact. Use genai.list_models() to see all available models for your account. Current valid model IDs:

Vertex AI: uses a different model path format — publishers/google/models/gemini-1.5-pro instead of the plain ID.

Streaming not working

Use stream=True and iterate over chunks — do not call response.text on the streaming response object:

for chunk in model.generate_content("Write a poem", stream=True):
    print(chunk.text, end="")

• Do not call response.text on a stream: calling response.text on a streaming response raises ValueError — use chunk.text inside the loop instead.
• Function calling + streaming: if you use tool/function calling with streaming, call response.resolve() after the loop before accessing tool_calls.