Newsletter image

Subscribe to the Newsletter

Join 10k+ people to get notified about new posts, news and tips.

Do not worry we don't spam!

By pressing the Subscribe button, you confirm that you have read and are agreeing to our Privacy Policy and Terms of Use

Search

GDPR Compliance

We use cookies to ensure you get the best experience on our website. By continuing to use our site, you accept our use of cookies, Privacy Policy, and Terms of Service.

xAI - Reasoning

xAI Releases Grok 3

Discover xAI's Grok 3, the latest AI model with enhanced reasoning and DeepSearch features. Explore subscription tiers for powerful AI tools today!
2025-02-18
Updated 2025-03-07 11:27:48

Grok 3 Unleashed: xAI’s Big Bet on AI Supremacy

On February 18, 2025, xAI dropped Grok 3, their latest AI heavyweight, into the ring, and the tech world’s buzzing like a kicked hive. Announced via a livestream on X with 100,000 viewers (catch it here), Elon Musk’s crew is calling it the “smartest AI on Earth,” built to outmuscle OpenAI’s ChatGPT, Google’s Gemini, and China’s DeepSeek. Trained on a monstrous 200,000 Nvidia H100 GPUs, it’s a flex of raw power—but is it genius or just another Musk hype bomb? Let’s tear it apart. Check the official xAI blog for the full announcement.

The Muscle: Colossus and a Family of Models

Grok 3 isn’t one trick pony—it’s a whole damn stable. You’ve got the flagship Grok 3, a leaner Grok 3 mini, and reasoning variants (Grok 3 Reasoning and mini Reasoning) that mimic human step-by-step logic. xAI’s Colossus supercomputer in Memphis, Tennessee—doubled from 100,000 to 200,000 H100 GPUs in 92 days—pumped 200 million GPU-hours into this beast. That’s 10 times Grok 2’s 20 million, per Musk’s X post. Compared to Grok 1’s 2023 debut on 10,000 GPUs, it’s a rocket leap. But raw compute’s just the engine—does it drive?

Benchmarks: Top Dog or Cherry-Picked Champ?

xAI’s waving shiny scorecards: Grok 3 and mini allegedly beat OpenAI’s GPT-4o, Gemini-2 Pro, Claude 3.5 Sonnet, and DeepSeek-V3 on select tests. Math? AIME 2025 gives Grok the edge over GPT-4o and DeepSeek’s R1. Science? GPQA (grad-level physics, bio, chem) nods its way. Coding? LCB benchmarks crown it. X’s Chatbot Arena has Grok 3 at #1, with users raving it’s “insane” and “wild.” But skeptics on X—like 

@AIOutsider

—call bullshit: where’s FrontierMath, Arc-AGI, or HLE scores? Without blind, brutal tests, it’s a flex with no spine. Numbers don’t lie, but cherry-picking does.

Features: DeepSearch, Reasoning, and a Paywall Catch

Grok 3’s loaded with tricks. DeepSearch, dubbed a “next-gen search engine,” trawls the web and X for sourced summaries—think OpenAI’s Deep Research with Musk’s edge. “Think” mode shows its brain live, crunching math, science, or code step-by-step, while “Big Brain” mode dials compute to 11 for heavy lifts. Voice mode’s teased for next week via the Grok app, promising synthesized banter. Want in? It’s locked behind X Premium+ ($40/month, up from $22—details here: 

https://x.com/premium

) or SuperGrok ($30/month via app and Grok.com). Free tier? Dead—Grok 2’s it now. X users are griping: “accessible” my ass. Scope the Grok webpage for the latest: https://grok.com.

The Drop: Livestream Swagger and a Beta Warning

The release hit X hard—100,000 tuned in (replay: https://t.co/HTK4u2aYRu) as Musk and xAI CEO Alex Vieira touted Grok 3’s “maximally truth-seeking” soul, less woke than rivals. It’s a beta, though—Musk admits “imperfections” they’ll patch daily. Grok 2’s going open-source “soon” (once Grok 3’s “mature”), but for now, it’s a pay-to-play toy. The livestream flexed Colossus’s build—122 days for the first 100,000 GPUs, 92 more to double it—matching Tesla’s Shanghai gigafactory pace. xAI’s $10 billion raise (valuing them at $75 billion, per Bloomberg) backs the brag, but delivery’s the proof. More from the xAI blog: https://x.ai/blog.

Competition: A Packed AI Cage Fight

Grok 3 lands in a brawl. OpenAI’s GPT-4o just laughed off Musk’s $97.4 billion buyout bid, DeepSeek’s R1 stunned with lean brilliance, and Google’s Gemini lurks. Israel’s IDF using OpenAI tech for Gaza strikes (AP, Feb 18) flips AI from chat to war—Grok’s got bigger shadows to dodge. Humane’s AI Pin crashed at $116 million (HP’s kill, Feb 18), a warning hype can tank. xAI’s banking on Grok 3’s power and “truth” angle—built with human feedback over canned morals—to stand out. Ballsy, but the ring’s crowded.

The Verdict: Crown or Clown Show?

Grok 3’s a beast—200,000 GPUs, top Arena rank, DeepSearch swagger. X’s split: fans scream “best ever,” doubters demand harder tests. It’s 1:42 AM CET, Feb 19, and the day’s young, but this is Musk’s loudest AI swing yet. Beats the EU’s Clean Deal hot air—xAI’s got silicon, not just speeches. Still, paywalls and beta bugs sting. Genius or gimmick? Shell out $40 (X Premium+: 

https://x.com/premium

) or hit Grok.com (https://grok.com) and judge—or wait for the smoke to clear. What’s the dumbest hype here, or the real win? I’m sniffing more as it breaks—stay tuned.

Links Added

Livestream Tweet: https://t.co/HTK4u2aYRu (Musk’s Feb 18 post linked in sources, e.g., Web ID 6).

Official Blog Announcements: https://x.ai/blog (xAI’s blog hub, covering Grok updates).

X Subscription: 

https://x.com/premium

 (X’s Premium page for subscription details).

Grok Webpage: https://grok.com (Direct link from xAI’s rollout, per Web ID 1).

Prev Article
Zonos TTS on Pinokio Computer
Next Article
Clean Deal, Unreal: EUs AI Spiel

Related to this topic: