Groq free API limits and failover

What is free on Groq

GroqCloud has a Free plan for building and testing, and a paid Developer plan with higher limits. Groq publishes separate Free plan limits per model. Some models, such as llama-3.3-70b-versatile and llama-3.1-8b-instant, are now enterprise-only and absent from the free table.

Examples listed as free when we checked: openai/gpt-oss-120b , openai/gpt-oss-20b , qwen/qwen3.8-27b , openai/gpt-oss-safeguard-20b , whisper-large-v3-turbo . Check the current catalog and your account access. Ficelle refreshes catalog data and applies eligibility checks during selection.

Listed free model examples on Groq

Qwen3.8 27B qwen/qwen3.8-27b
gpt-oss-120b openai/gpt-oss-120b

Groq free-tier limits

Limits apply per organization. The rate-limit page, rechecked on 6 October 2026, lists openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b at 30 requests per minute, 1,000 per day, 8,000 tokens per minute and 200,000 tokens per day each. openai/gpt-oss-safeguard-20b has lower limits: 5 requests per minute and 2,000 tokens per minute, with the same daily request and token caps. Whisper models get 20 requests per minute and 2,000 per day. Groq calls the table a high-level summary with exceptions; your exact limits are in the console. Cached tokens do not count.

Get a Groq API key

Create a GroqCloud account, then create a key in the console.

https://console.groq.com/keys

OpenAI compatibility

Groq describes its API as mostly compatible with OpenAI client libraries. logprobs, logit_bias, top_logprobs and messages[].name return a 400, n must be 1, and a temperature of 0 is converted to 1e-8.

base URL

https://api.groq.com/openai/v1

Data and privacy

Groq does not retain inference data by default. It may log inputs and outputs for up to 30 days to troubleshoot or investigate abuse, and every customer can turn on Zero Data Retention in Data Controls.

Through Ficelle, your Groq key stays on your machine and prompts go straight to Groq: there is no Ficelle proxy in between.

Before you rely on it

For a 429, inspect both request and token headers: x-ratelimit-remaining-requests refers to requests per day, and x-ratelimit-remaining-tokens to tokens per minute. Retry-After is present only on 429 responses.
The published table is a summary; the per-organization limits in your console are authoritative.
Popular Llama models moved to enterprise-only access.
Long prompts can exhaust TPM before RPM: Safeguard has 2,000 TPM; the listed GPT-OSS chat and Qwen models have 8,000 TPM.

Use Groq with automatic failover

Ficelle can include Groq behind your local OpenAI-compatible endpoint when the integration is enabled and its models pass free-access and capability checks. Groq is listed in Ficelle Pro's maintained provider pack. Check your installed configuration, credentials and account entitlement before relying on it. Another eligible candidate must be available for failover to succeed. Local routing guards do not control upstream billing when an account permits paid usage beyond its free allocation. See strict-zero routing and account limits.

shell

# Groq ships in Ficelle Pro's maintained provider pack.
ficelle doctor --text          # check this install before configuring provider access

python

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8646/v1", api_key="ficelle-local")

# Ficelle selects among configured, enabled, eligible models.
# If none can serve the request, it returns an error.
client.chat.completions.create(
    model="ficelle/auto-tools",
    messages=[{"role": "user", "content": "..."}],
)

See what Ficelle Pro includes →

Pick a model type

Choose a model type for your request. Eligibility depends on the installed provider configuration, catalog data, free-access evidence and capability checks. A catalog listing alone does not confirm that your account can call a model.

Questions

Is the Groq API free?

GroqCloud has a Free plan for building and testing, and a paid Developer plan with higher limits. Groq publishes separate Free plan limits per model. Some models, such as llama-3.3-70b-versatile and llama-3.1-8b-instant, are now enterprise-only and absent from the free table.

What are Groq's free rate limits?

Limits apply per organization. The rate-limit page, rechecked on 6 October 2026, lists openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b at 30 requests per minute, 1,000 per day, 8,000 tokens per minute and 200,000 tokens per day each. openai/gpt-oss-safeguard-20b has lower limits: 5 requests per minute and 2,000 tokens per minute, with the same daily request and token caps. Whisper models get 20 requests per minute and 2,000 per day. Groq calls the table a high-level summary with exceptions; your exact limits are in the console. Cached tokens do not count.

Is Groq OpenAI-compatible?

Partly. Groq describes its API as mostly compatible with OpenAI client libraries. logprobs, logit_bias, top_logprobs and messages[].name return a 400, n must be 1, and a temperature of 0 is converted to 1e-8. Base URL: https://api.groq.com/openai/v1.

What happens when Groq rate-limits my agent?

Ficelle applies a cooldown to the affected model or provider quota scope and can retry another configured, enabled candidate that meets the request requirements and free-access checks. Recovery probes can restore eligibility. If no eligible candidate can serve the request, Ficelle returns an error; it does not switch to a paid fallback.

Sources

Official Groq references. Last complete source review: 28 September 2026; later partial updates are dated in the text: