Groq vs Grok: One Is an AI Inference Cloud, One Is Elon's Chatbot (2026)

The 30-Second Answer

If you are looking for fast inference hardware or an API that serves open models cheaply, you want Groq (with a q). It designed the LPU chip and runs GroqCloud, which serves GPT-OSS, Qwen, and Llama models at 450 to 1,000 tokens per second. Jonathan Ross, who started Google's TPU project, founded it in 2016.

If you are looking for the AI assistant on X, or an API to call xAI's models, you want Grok (with a k). It is a chatbot and model family from Elon Musk's xAI, launched in late 2023. The current flagship is grok-4.7. The API also lists grok-build-0.1 for agentic coding and grok-4.3 for 1M-token prompts.

The confusion is not your fault. The names are phonetically identical, both arrived during the same AI boom, and Groq had the name and the trademark seven years before Grok existed. Groq's CEO said as much in a public cease-and-desist letter, which we cover below.

Groq vs Grok, Side by Side

The fastest way to tell them apart is to line up the facts. Almost nothing overlaps except the sound of the name.

The single biggest tell: Groq does not have a chatbot you talk to, and Grok does not sell a general inference cloud for other companies' models. If someone says "I ran it on Groq," they served a model. If someone says "I asked Grok," they chatted with xAI.

Groq (with a q): The Inference Cloud

Groq started as a chip company and now calls itself an inference cloud. Its chip, the LPU (Language Processing Unit), is built for one job: running trained models fast. The architecture is deterministic and low-latency, which sidesteps the memory-bandwidth wall that GPUs hit during token generation. GroqCloud's model list puts GPT-OSS 20B at about 1,000 tokens per second, GPT-OSS 120B at about 500, and Qwen3.8-27B at about 450.

Jonathan Ross founded Groq in 2016 with Douglas Wightman, both former Google engineers. Ross had started Google's TPU project as a 20% initiative, so the LPU comes from the same lineage of purpose-built inference silicon. The chip began life as the Tensor Streaming Processor and was rebranded to LPU after the LLM boom made language inference the dominant workload.

The company changed shape in December 2025. Groq signed a non-exclusive licensing agreement that gave NVIDIA its inference technology. Founder Jonathan Ross, president Sunny Madra, and other Groq staff joined NVIDIA, Simon Edwards became CEO, and GroqCloud kept running. NVIDIA then announced its LPX platform built on Groq's technology at GTC 2026.

Groq refocused on the cloud. It raised $650M in June 2026, led by Disruptive and Infinitum. It joined the NVIDIA Cloud Partner program on August 12, 2026, and closed a $350M Series A on August 17 at a $3.5B valuation, with planned participation from NVIDIA. Groq reports 13 data centers across North America, Europe, the Middle East, and Asia Pacific, and plans to grow from 54 MW to more than 200 MW in 2027. On August 24 it said it will be among the first clouds to deploy NVIDIA Groq 3 LPX with Vera Rubin NVL72, citing 3,400 output tokens per second on Gemma 4 31B at 100K context in Artificial Analysis testing.

Groq is one of several inference clouds that serve open models through an OpenAI-compatible API. Morph is another, running open-weight coding models on kernels tuned for code generation. See Morph Open Source Models and Fireworks AI pricing for how the inference-provider market compares.

Grok (with a k): xAI's Chatbot

Grok is the assistant from xAI, Elon Musk's AI company. It launched in November 2023, took its name and personality from the Hitchhiker's Guide to the Galaxy, and is wired into X so it can answer with real-time posts. Grok is both a consumer chatbot and an API-accessible model family.

As of October 2026, xAI's docs name grok-4.7 as the flagship for code and everything else, with configurable reasoning and a May 2026 knowledge cutoff. grok-build-0.1 covers high-volume agentic coding, grok-4.3 offers a 1M-token context, and grok-4.20-0309 covers reasoning, non-reasoning, and multi-agent modes. Consumers reach Grok through X, the Grok apps, and the SuperGrok subscription.

Every rate doubles on prompts of 200k tokens or more. Cached input runs $0.20 to $0.50 per M. The US regional endpoint bills 1.1x the global rate. Full breakdown on our Grok API pricing page. Verify current numbers on docs.x.ai before relying on them.

The xAI API is OpenAI-compatible: point the OpenAI SDK at https://api.x.ai/v1, pass your xAI key, and set the model to grok-4.7 or grok-build-0.1. So Grok (the model) and Groq (the hardware that could, in principle, serve open models) are reachable through similar-looking OpenAI-shaped code, which does nothing to reduce the confusion.

GroqCloud Pricing and Groq Models

GroqCloud charges per token with no idle infrastructure fee. It serves open models, so the prices track model size rather than a single flagship. The lineup shrank in 2026. Llama 4 Scout and Qwen3 32B are gone, and the two Llama models moved to Enterprise pricing. These are the Groq API rates listed in the GroqCloud docs in October 2026.

Speech-to-text is billed per audio hour. Whisper Large V3 costs $0.111 per hour and Whisper Large V3 Turbo costs $0.04 per hour. The live model list is at https://api.groq.com/openai/v1/models.

Two discounts matter. The Batch API costs 50% less than synchronous calls for jobs with a 24-hour to 7-day window. Prompt caching cuts the cost of cached tokens by 50%, applies automatically, and expires after 2 hours without use. The Free plan caps GPT-OSS 120B at 30 requests per minute and 200,000 tokens per day. The paid Developer plan raises those limits and adds Batch and Flex processing. Because GroqCloud serves open weights, the comparison that matters is provider-to-provider on the same model.

Grok API Pricing

Grok is a closed model family, so xAI publishes one price sheet for its own models. The flagship grok-4.7 sits at the top. grok-4.3 and grok-build-0.1 are cheaper per token for higher-volume work.

Every rate doubles once a prompt reaches 200k tokens, so grok-4.3 becomes $2.50/$5.00 per M on large-context requests. Cached input runs $0.20 to $0.50 per M depending on the model. For the full model-by-model breakdown and the long-context tier math, see Grok API pricing, and for how xAI's rates sit against other providers, see the LLM API pricing overview.

The Naming Collision: q Came First

Groq registered its trademark in 2016, seven years before xAI existed. When Elon Musk announced the Grok chatbot in November 2023, Groq's CEO Jonathan Ross responded with a public cease-and-desist blog post, dated November 30, 2023. It is a real letter, and it is funny on purpose.

The letter goes on in the same register. Ross wrote that he could see why xAI might want the name ("You like fast things: rockets, hyperloops, one-letter company names"), recommended Musk rename the bot to Slartibartfast (a nod to the same Hitchhiker's Guide source Grok drew from), and joked about being asked about the snarky chatbot by his great aunt over Thanksgiving dinner. Under the comedy is a straight trademark claim: Groq was first, both operate in AI, and the phonetic overlap is exactly the kind of thing that causes consumer confusion.

The trademark friction did not stop at a blog post. The USPTO reportedly rejected xAI's application to register Grok, citing similarity to Groq and to another company, Grokstream, and a separate startup has claimed rights to the Grok name in the SaaS category. The upshot for you as a reader: the two products still coexist, the search results still collide, and the safest habit is to check the consonant. q is the chip, k is the chatbot.

Which One Do You Want?

If your task is coding and you care about the inference layer, the useful frame is not Groq versus Grok, it is which provider serves the model you want, at what speed, at what price. Morph serves open coding models (Kimi K3, GLM-5.3, DeepSeek V4.1 Flash) on custom kernels through an OpenAI-compatible API, the same category Groq competes in. See Morph Open Source Models.

Frequently Asked Questions

Are Groq and Grok the same thing?

No. Groq (with a q) is an AI hardware company founded in 2016 that builds LPU inference chips and runs GroqCloud. Grok (with a k) is a chatbot and model family from Elon Musk's xAI, launched in 2023 and built into X. Separate companies, unrelated products, one consonant apart.

Is Groq owned by Elon Musk?

No. Groq was founded by Jonathan Ross and Douglas Wightman, former Google engineers. Its 2026 rounds were led by Disruptive and Infinitum, with NVIDIA planning to join the August Series A. Grok the chatbot is the one owned by Musk, through xAI. The name overlap is exactly why this gets asked.

What is the difference between an LPU and a GPU?

Groq's LPU (Language Processing Unit) is designed only for LLM inference, with a deterministic, low-latency architecture that avoids the memory-bandwidth wall GPUs hit on token generation. GroqCloud lists GPT-OSS 20B at about 1,000 tokens per second. GPUs stay more general-purpose and dominate training. NVIDIA licensed the LPU design in December 2025 and pairs it with Vera Rubin GPUs as Groq 3 LPX.

How much does GroqCloud cost?

Per token, no idle fee. As of October 2026, GPT-OSS 20B is $0.075/$0.30 per M, GPT-OSS 120B is $0.15/$0.60, and Qwen3.8-27B (preview) is $0.80/$4.00. Both Llama models are Enterprise contact-sales. The Batch API saves 50%, and cached tokens cost 50% less.

What models does Groq support?

Production models are GPT-OSS 120B, GPT-OSS 20B, Llama 3.1 8B Instant, Llama 3.3 70B Versatile, and two Whisper speech-to-text models. Preview models include Qwen3.8-27B, MiniMax M2.7, GPT-OSS Safeguard 20B, Llama Prompt Guard 2, and Orpheus text-to-speech.

Does Groq have a free tier?

Yes. The Free plan has per-model rate limits. GPT-OSS 120B gets 30 requests per minute, 1,000 requests per day, and 200,000 tokens per day. The Developer plan raises the limits and adds Batch and Flex processing.

How much does the Grok API cost?

As of October 2026, grok-4.7 is $2.00/$6.00 per M on a 500k context, grok-4.3 is $1.25/$2.50 per M on a 1M context, and grok-build-0.1 is $1.00/$2.00 per M on 256k. Every rate doubles on prompts of 200k tokens or more. Full detail on our Grok API pricing page.

Did Groq sue xAI over the Grok name?

Groq did not file a public lawsuit, but it did send xAI a cease-and-desist. On November 30, 2023, CEO Jonathan Ross published a humorous public letter noting the phonetic overlap and Groq's earlier trademark. The USPTO reportedly rejected xAI's Grok trademark application, citing similarity to Groq and to Grokstream.