You are seeing the message “Our servers are experiencing high traffic right now, please try again in a minute” when calling AI models. This points to either a provider outage or your requests hitting platform limits. Here is the quickest path to a fix.
Fix the high traffic error for AI model calls
This issue presents as intermittent or constant failures from chat or API calls with the exact message above. In recent incidents, users reported that it impacted multiple models at once including Claude, Sonnet, Pro, and non Pro tiers, which suggests a platform level event rather than a single model fault.
Read More:Stitch project not loading under Antigravity
Solution Overview
Aspect
Detail
Root Cause
Global service outage or your project reached concurrency and rate limits
Primary Fix
Confirm service status, check usage limits, then retry with exponential backoff or switch model or region
Complexity
Easy
Estimated Time
5 to 20 minutes
Fix it now
Step by step solution
Step 1. Check current service status
Open the public status page for Antigravity. If it shows an incident, waiting is the only action until recovery completes.
In your workspace open Advanced Settings then Model and review any limits or caps that you configured. If you recently increased traffic, you may be rate limited.
If you are on Google Cloud Vertex AI, also check quotas in the Console under Vertex AI Quotas or use the CLI:
– View project and region to confirm you are inspecting the right quotas
– gcloud config list
– Open quotas in Console
– https://console.cloud.google.com/iam-admin/quotas
If you see errors close to your limits increase quota requests in Console.
Step 3. Add retry with exponential backoff now
Even during partial incidents, retries with jitter often succeed.
Bash example with curl and backoff
API_KEY="YOUR_API_KEY"
URL="https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-latest:generateContent?key=${API_KEY}"
body='{
"contents": [{"parts":[{"text":"Say hello"}]}]
}'
max=6
n=0
sleep_sec=1
while true; do
status=$(curl -s -o resp.json -w "%{http_code}" \
-H "Content-Type: application/json" \
-d "${body}" "${URL}")
if [ "$status" -ge 200 ] && [ "$status" -lt 300 ]; then
echo "Success"
cat resp.json
break
fi
if [ "$n" -ge "$max" ]; then
echo "Failed after retries with status $status"
cat resp.json
exit 1
fi
# jitter
jitter=$(( (RANDOM % 1000) + 1 ))
sleep_time=$(awk "BEGIN {printf \"%.2f\", ${sleep_sec} + ${jitter}/1000}")
echo "Retry $((n+1)) in ${sleep_time}s status $status"
sleep "${sleep_time}"
sleep_sec=$(( sleep_sec * 2 ))
n=$(( n + 1 ))
done
Python example with exponential backoff and jitter
import os, time, random, json, requests
API_KEY = os.getenv("API_KEY")
URL = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-latest:generateContent?key={API_KEY}"
body = {
"contents": [{"parts":[{"text":"Say hello"}]}]
}
def call_with_retries(max_attempts=6):
delay = 1.0
for attempt in range(1, max_attempts + 1):
r = requests.post(URL, json=body, timeout=30)
if 200 <= r.status_code < 300:
return r.json()
if attempt == max_attempts:
raise RuntimeError(f"Failed after {attempt} attempts code={r.status_code} body={r.text}")
jitter = random.uniform(0, 1)
sleep_for = delay + jitter
print(f"Attempt {attempt} failed code={r.status_code} retrying in {sleep_for:.2f}s")
time.sleep(sleep_for)
delay *= 2
print(json.dumps(call_with_retries(), indent=2))
Step 4. Lower concurrency and batch your requests
Temporarily reduce parallel calls and batch prompts where possible.
Many outages are partial. A smaller sustained rate with backoff frequently succeeds.
Step 5. Switch model or region
If status is green and your quota is fine, switch the model or its region. Use a lighter model for now.
Example switching to flash model endpoint
# previously using gemini-1.5-pro-latest
URL="https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-latest:generateContent?key=${API_KEY}"
Vertex AI example switching region for text generation
Always implement exponential backoff with jitter for 429 and 5xx responses.
Right size concurrency per model and region and review quotas monthly.
Isolate workloads per project and region to avoid a single point of failure.
Add a kill switch in config to change model or region without code deploys.
Monitor error budgets and track model error rate as a key service level indicator.
Final Thought
This error usually traces to a service incident or account limits. By checking status, confirming quotas, adding robust retries, and switching models or regions, you can restore stability fast and prevent repeats.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
