Groq
Cloud AI inference platform serving open models on custom LPU hardware.
HN commenters use Groq for very fast, low-cost inference12345 on open models like Llama 3, gpt-oss-20b and Whisper.678 Complaints are rare; the one caveat raised is a worry that the service may change,9 since its policy of not storing data by default has recently ended.10
Why people recommend it
Speed · 3
Inference is fast enough for latency-sensitive pipelines and even system loops.
“With Groq speed (Llama 3 @ 800t/s)”
“We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.”
Show 1 more
“I used Groq (not to be confused with Grok), because it is free and very fast.”
Model choice · 3
Commenters run several open models on it, including Llama 3, gpt-oss-20b and whisper-large-v3-turbo.
“With Groq speed (Llama 3 @ 800t/s)”
“We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.”
Show 1 more
“It's incredibly cheap and works reliably for me.”
Low cost · 3
Described as cheap, and as free for one user's scripting use.
“It's incredibly cheap and works reliably for me.”
“I used Groq (not to be confused with Grok), because it is free and very fast.”
Show 1 more
“I'm a fan, and I use Groq a lot for systems I build.”
Data privacy · 2
Suggested as an inference-only option for people worried about prompts being used for training.
“you can use an inference only solution like Groq.”
“I'm a fan, and I use Groq a lot for systems I build.”
Reliability · 1
One user says it works reliably for voice transcription.
“It's incredibly cheap and works reliably for me.”
Caveats and dissent
Policy changes · 1
One fan worries about the impact of recent changes, noting its no-data-storage default has ended.
“I'm a fan, and I use Groq a lot for systems I build.”
Name confusion · 1
Easily confused with Grok, a different product.
“I used Groq (not to be confused with Grok), because it is free and very fast.”
Who recommends it in their own words
Threads where it came up
- Ask HN: Senior software engineers, how do you use Claude Code? 1 mention
- Show HN: Replacing my OS process scheduler with an LLM 1 mention
- Nvidia's $20B Antitrust Loophole (Not an Acquisition) 1 mention
- Using Vectorize to build an unreasonably good search engine in 160 lines of code 1 mention
- GLM-4.7: Advancing the Coding Capability 1 mention
- Ask HN: How can I get better at using AI for programming? 1 mention
Data: groq.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 7
“you can use an inference only solution like Groq.”
“With Groq speed (Llama 3 @ 800t/s)”
“I'm a fan, and I use Groq a lot for systems I build.”
“We use Groq with gpt-oss-20b, which gives great results and only adds ~250ms to the processing pipeline.”
“(or Groq with other models)”
“It's incredibly cheap and works reliably for me.”
“I used Groq (not to be confused with Grok), because it is free and very fast.”