Read the error message
Groq's 429 message names the model, the organization, the service tier, the limit type and the wait time. This example was posted in a public bug report:
Example 429 body text
Rate limit reached for model openai/gpt-oss-120b in organization org_<id>
service tier on_demand on tokens per day (TPD): Limit 200000, Used 193487,
Requested 7464. Please try again in 6m50.832s.| Limit in the message | What ran out | What to do |
|---|---|---|
| requests per minute (RPM) | Too many requests in 60 seconds | Wait a few seconds and retry |
| tokens per minute (TPM) | Too many tokens in 60 seconds | Wait, or send shorter prompts |
| requests per day (RPD) | The daily request quota | Wait the time in the message, or use another model |
| tokens per day (TPD) | The daily token quota | Wait the time in the message, or use another model |
A different error is HTTP 413 with "Request too large". It means that one request is larger than the TPM limit of the model. Waiting does not help. Make the request smaller, or use a model with a higher TPM limit.
Groq free plan limits
Groq applies rate limits to the organization, not to each user or key. These values are from the Free Plan tab of Groq's rate limits page. Cached tokens do not count toward the limits.
| Model | RPM | RPD | TPM | TPD |
|---|---|---|---|---|
openai/gpt-oss-120b | 30 | 1,000 | 8,000 | 200,000 |
openai/gpt-oss-20b | 30 | 1,000 | 8,000 | 200,000 |
openai/gpt-oss-safeguard-20b | 3 | 1,000 | 2,000 | 200,000 |
qwen/qwen3.8-27b | 30 | 1,000 | 8,000 | 200,000 |
meta-llama/llama-prompt-guard-2-86m | 30 | 14,400 | 15,000 | 500,000 |
whisper-large-v3 | 20 | 2,000 | - | - |
Whisper models also have audio limits: 7,200 audio seconds per hour and 28,800 per day. The table below shows the Groq rows in the FreeLLMAPI catalog. If a catalog value and Groq's page differ, Groq's page is the authority.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| Qwen3.8 27B | Groq | 262K | 30 req/min 250 req/day 70,000 tokens/min ~6M | Yes | Yes |
| GPT-OSS 120B | Groq | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| GPT-OSS 20B | Groq | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| GPT-OSS Safeguard 20B | Groq | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| ALLaM 2 7B | Groq | 4K | 30 req/min 1,000 req/day 6,000 tokens/min 500,000 tokens/day ~15M | No | No |
| Llama Prompt Guard 2 22M | Groq | 512 | 30 req/min 14,400 req/day 15,000 tokens/min 14.4K rpd | No | No |
| Llama Prompt Guard 2 86M | Groq | 512 | 30 req/min 14,400 req/day 15,000 tokens/min 14.4K rpd | No | No |
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
The rate limit headers
Each Groq response has rate limit headers. Read them to slow down before you get a 429.
| Header | Refers to | Meaning |
|---|---|---|
x-ratelimit-limit-requests | RPD | Requests allowed per day |
x-ratelimit-remaining-requests | RPD | Requests left today |
x-ratelimit-reset-requests | RPD | Time until the request limit resets, for example 2m59.56s |
x-ratelimit-limit-tokens | TPM | Tokens allowed per minute |
x-ratelimit-remaining-tokens | TPM | Tokens left this minute |
x-ratelimit-reset-tokens | TPM | Time until the token limit resets, for example 7.66s |
retry-after | Any | Seconds to wait. Groq sets it only after you hit a limit. |
See the headers
curl -s -D - -o /dev/null https://api.groq.com/openai/v1/chat/completions \
-H "Authorization: Bearer <groq-key>" \
-H "Content-Type: application/json" \
-d '{"model": "openai/gpt-oss-20b", "messages": [{"role": "user", "content": "hi"}]}' \
| grep -i -E 'x-ratelimit|retry-after'Retry correctly
Use retry-after when Groq sends it. If it is missing, use exponential backoff with a cap. Do not retry daily limits in a tight loop. Each retry is a new request.
Retry with retry-after (Python)
import time
from openai import OpenAI, RateLimitError
client = OpenAI(base_url="https://api.groq.com/openai/v1", api_key="<groq-key>")
def ask(prompt, model="openai/gpt-oss-20b"):
for attempt in range(5):
try:
r = client.chat.completions.create(model=model, messages=[{"role": "user", "content": prompt}])
return r.choices[0].message.content
except RateLimitError as e:
wait = e.response.headers.get("retry-after")
time.sleep(float(wait) if wait else min(60, 2 ** attempt))
raise RuntimeError("Groq rate limit: retries used up")- If the message names RPD or TPD and the wait is minutes, change to another model or provider.
- To get higher limits from Groq, upgrade to the Developer tier. It needs a payment method.
- HTTP 498 is a different error. It means that the flex tier is at capacity.
Plan your Groq budget
The free limits are small when you do the arithmetic. Check which limit stops you first for your workload.
- Requests: at 30 requests per minute, 1,000 requests per day last about 34 minutes of full-speed use.
- Tokens per minute: 8,000 TPM allows about four requests of 2,000 tokens each minute. One request of 9,000 tokens gets error 413.
- Tokens per day: 200,000 TPD is about 25 requests of 8,000 tokens. Coding agents send large prompts, so TPD often runs out before RPD.
- Organization: all keys and all members of the organization share these numbers.
For agent work, keep prompts short and clear old context often. Use Groq for fast, short calls. Send long-context calls to providers with a higher token limit.
Fail over when Groq is full
FreeLLMAPI is an open-source gateway that runs on your computer. It keeps RPM, RPD, TPM and TPD counters for each Groq model and key. It checks them before each request. When Groq is full, the router sends the request to the next free provider.
- A short 429 on a minute limit puts the key on a 90 second cooldown.
- A 429 on a daily limit uses a longer cooldown, or Groq's retry time if it is longer.
- When a Groq error body states the real limit, the router saves it. For example, a 413 with
Limit 30000, Requested 33476sets the TPM limit to 30,000. - The
X-Fallback-Trailheader shows each failed hop, for examplegroq/llama-3.3-70b key1=rate_limited.
Call the gateway instead of Groq
curl -i http://localhost:3001/v1/chat/completions \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "auto:fast", "messages": [{"role": "user", "content": "hi"}]}'auto:fast prefers the models with the best measured speed. When Groq is at its limit, other providers answer. They can be slower than Groq.
Frequently asked questions
What does 'Rate limit reached for model' mean on Groq?
Your organization used all of one limit for that model. The message names the limit (RPM, RPD, TPM or TPD) and the time to wait.
What are the Groq free tier rate limits?
They differ by model. On the free plan, openai/gpt-oss-120b allows 30 requests per minute, 1,000 per day, 8,000 tokens per minute and 200,000 tokens per day.
Why do I get Groq error 413 Request too large?
One request has more tokens than the TPM limit of the model. Make the prompt shorter or use a model with a higher TPM limit. Waiting does not fix it.
Does a new Groq API key reset the rate limit?
No. Groq applies limits to the organization, so all keys in it share the same limits.
How do I increase my Groq rate limit?
Upgrade to the Developer tier with a payment method. To stay free, add other providers behind a gateway that fails over when Groq is full.
Related guides
- Free LLM API rate limits
- free groq api
- Gemini API quota exceeded (429 RESOURCE_EXHAUSTED)
- NVIDIA NIM API rate limit
- Groq vs Cerebras
Sources
- GroqCloud: Rate limits (checked 20 November 2026)
- GroqCloud: Errors (checked 20 November 2026)
- GroqCloud: Billing FAQs (checked 20 November 2026)
- Public report of the Groq TPD message (Discourse) (checked 20 November 2026)
- FreeLLMAPI docs: Quota and cooldown engine (checked 20 November 2026)
- FreeLLMAPI docs: Response headers (checked 20 November 2026)