Gemini API Rate Limits Complete Guide: RPM, TPM, 429 र Tier Upgrade

Gemini API मा 429 प्रायः code bug होइन, multi-dimensional rate limit हो। RPM requests/minute, TPM tokens/minute, RPD requests/day र IPM images/minute हुन्। कुनै dimension limit नाघे request reject हुन सक्छ। Exact values model, tier, project र current official row मा निर्भर हुन्छन्; static table स्थायी contract होइन।

चार limits

• RPM: हरेक API call एक request; burst मा पहिले सकिन सक्छ।
• TPM: prompt/context/output tokens; ठूलो document मा RPM भन्दा पहिले लाग्न सक्छ।
• RPD: daily total; repeated test ले छिटो भरिन सक्छ।
• IPM: image generation को छुट्टै limit।

Limits Google Cloud project मा हुन्, key मा होइन; एउटै project का keys एउटै pool share गर्छन्। Exact model row र AI Studio rate-limit page हेर्नुहोस्। Free learning/prototype का लागि, Tier 1/2/3 production scale का लागि हो; current billing/eligibility verify गर्नुहोस्।

429 र backoff

Response मा 429, RESOURCE_EXHAUSTED र कहिलेकाहीँ retryDelay आउन सक्छ। 429 मात्र retry गर्नुहोस्; अरू errors surface गर्नुहोस्।

import time, random
import google.generativeai as genai
def call_with_retry(prompt, max_retries=5):
    genai.configure(api_key="YOUR_GPT88_API_KEY")
    model = genai.GenerativeModel("gemini-2.5-flash")
    for attempt in range(max_retries):
        try: return model.generate_content(prompt).text
        except Exception as exc:
            if "429" not in str(exc) and "RESOURCE_EXHAUSTED" not in str(exc): raise
            if attempt == max_retries - 1: raise
            delay = min(2 ** attempt, 32); time.sleep(delay + random.uniform(0, delay * .1))

Exponential backoff, jitter, max attempts र total time budget राख्नुहोस्। Headers बाट remaining requests/tokens/reset time पढेर console metrics सँग cross-check गर्नुहोस्। Token bucket/queue मा 20–30% margin, 80% alert, cache, Context Caching, admission control र fallback राख्नुहोस्। नयाँ key ले same-project quota बढाउँदैन; paid tier ले सबै limit हटाउँदैन।

Further Reading

• Error Code Reference