Short answer: Groq documentation describes DeepSeek-R1 distilled models, including deepseek-r1-distill-llama-70b and deepseek-r1-distill-qwen-32b. They are not the original DeepSeek-R1. As of November 10, 2026, I could not confirm from the official public documentation retrieved for this update whether either ID remains active or has been deprecated, or whether the 70B model is classified as Production or Preview. The Qwen model card excerpt labels that model Preview, but that alone does not confirm its current lifecycle status. Check the exact ID and its classification before relying on either model. (Deprecations · Supported Models)
What does “Groq DeepSeek” mean?
The name can refer to more than one Groq-documented model:
deepseek-r1-distill-llama-70b: Groq describes this as a 70-billion-parameter model built on the Llama-3.3-70B-Instruct framework and fine-tuned using outputs from DeepSeek-R1. (70B model card)deepseek-r1-distill-qwen-32b: Groq describes this as a 32-billion-parameter model built on Qwen-2.5-32B and distilled using DeepSeek-R1 outputs. The supplied model-card excerpt labels it Preview. (32B model card)
“Distilled” matters: Groq’s descriptions do not say these are the original DeepSeek-R1 or that they are equivalent to it. They are separately described models based on Llama and Qwen. Also, Groq and Grok are different names; a Grok comparison is not evidence about Groq’s API.
Check whether the exact model is active
Groq documents an authenticated Models API request for listing active model IDs. Set GROQ_API_KEY to your key, then run:
curl -sS https://api.groq.com/openai/v1/models \
-H "Authorization: Bearer $GROQ_API_KEY" \
-H "Content-Type: application/json"
Look for the exact ID you intend to call. Groq says this endpoint returns active models, but an ID in that list alone does not establish production classification. Check the current model listing and deprecation documentation as well. Groq’s catalog says Preview models are for evaluation and should not be used in production; the supplied Qwen card excerpt labels that model Preview. I could not confirm a current, exact-ID lifecycle result for either model from the public documentation retrieved for this update. The 70B model’s documented code example is not proof of current active status or production eligibility. (Supported Models · Deprecations)
Capabilities and evidence limits
For the 70B model, Groq lists mathematical problem-solving, coding assistance, and logical reasoning. Its page reports AIME 2024 Pass@1 of 70.0, MATH-500 Pass@1 of 94.5, and a CodeForces rating of 1,633. These are Groq-reported figures: the model page does not provide their methodology, evaluation date, or independent validation. They can suggest tasks to test, but do not establish that the model beats another option or will perform well on your workload. (70B model card)
The supplied 32B Preview-card excerpt lists a 128K-token context window, a 16,384-token maximum output, and prices of $0.69 per 1 million input tokens and $0.69 per 1 million output tokens. Those are the terms reported for that model card; do not apply them to the 70B model or treat Preview status as production approval. The 70B page does not provide a model-specific price or operating limits in the captured documentation. Confirm current prices, rate limits, and token limits for the exact model and plan before estimating costs. (32B model card)
Groq’s 70B page recommends a temperature from 0.5 to 0.7, ideally 0.6, and putting instructions in the user prompt rather than a system prompt. Treat this as that page’s guidance, not a universal rule; test prompt formats against your own evaluation set. (70B model card)
Python API call
Groq’s 70B model page shows a chat-completions call for that ID. Its Quickstart explains creating an API key and setting it as GROQ_API_KEY; the example below combines those documented steps. First check the active model list and confirm access to the exact ID. This illustrates the documented call pattern; it is not a claim that the request was executed or that the model is currently active or production-ready. (70B model card · Quickstart)
Install the SDK and set the key in your shell:
pip install groq
export GROQ_API_KEY="your-api-key"
Then call the model:
import os
from groq import Groq
client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
completion = client.chat.completions.create(
model="deepseek-r1-distill-llama-70b",
messages=[
{"role": "user", "content": "Solve this problem and show a concise explanation: ..."}
],
)
print(completion.choices[0].message.content)
If evaluating the 32B model, use deepseek-r1-distill-qwen-32b only after checking its current status and access; its supplied card excerpt labels it Preview, which is not a production recommendation. (32B model card)
Practical selection rule
Treat the model cards as descriptions, not proof that either ID is currently available for evaluation. First check the active-model endpoint and current lifecycle documentation for the exact ID. For production, confirm its classification, current pricing and limits, then test representative prompts against your quality, latency, and cost requirements. Groq-reported benchmark scores can help choose what to test, but they are not a substitute for those checks.
Sources
DeepSeek-R1-Distill-Llama-70B — GroqDocs