Hey there, fellow Grok explorers! Whether it was cracking math puzzles, spinning tales, or hallucinating brick-throwing basketball stars, Grok’s been a ride. But with all its fancy terms—AIME, DeepSearch, Big Brain Mode—it’s easy to get lost in the jargon. That’s where this glossary comes in.
Think of it as your trusty map to Grok’s universe. For the tech wizards, we’ve got the nitty-gritty: scores, features, and what makes Grok tick, straight from xAI’s playbook. Not a techie? No sweat—we’ve added simple, everyday explanations too, like calling Colossus a “mega-smart factory” instead of a GPU beast. Whether you’re here to geek out or just curious about this quirky AI, we’ve got you covered.
So, grab a coffee, and dive into the Grok Glossary. It’s the cheat sheet to everything we’ve learned—and a peek at what’s next for xAI’s brainchild. Let’s grok it together!
AIME (American Invitational Mathematics Examination)
- Tech Definition: A rigorous math competition for high school students. Grok 3 scored 52% on AIME 2024 and an impressive 93.3% with Think Mode on AIME 2025, highlighting its advanced problem-solving. Grok 4 Heavy pushed it further, achieving 100% on AIME 2025, setting new standards in mathematical reasoning.
- Simple Definition: A tough math showdown for teens—like the Olympics for brains. Grok 3 nailed it, and Grok 4 aced it completely, proving it’s a total math genius.
ARC-AGI (ARC-AGI Benchmark)
- Tech Definition: A challenging test for artificial general intelligence, measuring core knowledge and generalization. Grok 4 scored 15.9% on ARC-AGI V2 in July 2025, nearly doubling competitors and showcasing improved reasoning.
- Simple Definition: A puzzle test for “true” AI smarts—like figuring out patterns without hints. Grok 4 did way better than others, showing it’s getting closer to human-like thinking.
Aurora
- Tech Definition: Grok’s current image-generation model (replacing Flux in 2025), capable of creating detailed visuals from text prompts with enhanced realism and creativity, integrated into Grok’s multimodal features.
- Simple Definition: Grok’s picture-creator tool—like a digital artist that draws anything you describe, but better and less restricted than the old one.
Big Brain Mode
- Tech Definition: A competitive leaderboard for conversational AIs. Grok 3 scored 1402 points, excelling in dialogue and reasoning. Grok 4 improved further, ranking top-3 across categories like math and coding in LMSYS benchmarks.
- Simple Definition: Grok’s secret “supercharge” option—like upgrading to a massive brain for tackling the weirdest, hardest problems (but it’s not out for everyone yet).
Chatbot Arena
- Tech Definition: A competitive leaderboard for conversational AIs. Grok 3 scored 1402 points, excelling in dialogue and reasoning. Grok 4 improved further, ranking top-3 across categories like math and coding in LMSYS benchmarks.
- Simple Definition: A ranking contest for talkative AIs—Grok 3 got a high score for chatting and thinking smartly, and Grok 4 climbed even higher.
Colossus
- Tech Definition: xAI’s massive supercomputer with over 200,000 Nvidia H100 GPUs (expanding to H200s), which trained Grok 3 in 122 days and powered Grok 4’s development for unprecedented scale and speed.
- Simple Definition: A colossal, lightning-fast computer cluster—like a gigantic factory that builds and trains Grok models in record time.
DeepSearch
- Tech Definition: Grok’s real-time search integration that pulls fresh data from the web and X (formerly Twitter), synthesizing it into comprehensive answers. Ideal for current events or research, though accuracy dips on obscure topics due to source variability.
- Simple Definition: Grok’s instant online lookup tool—like a super-quick detective scanning the internet and X for up-to-date info, but it might miss rare details.k.
Flux
- Tech Definition: Grok’s temporary 2024 image-generation model, known for relaxed content rules (e.g., generating edgy images like armed cartoon characters), before being succeeded by the more refined Aurora in 2025.
- Simple Definition: Grok’s former picture-maker that allowed wild, unrestricted creations—like drawing cartoons in crazy scenarios—now upgraded to something better.
GPQA (Graduate-Level Google-Proof Q&A)
- Tech Definition: A PhD-level benchmark for scientific knowledge. Grok 3 scored 75%, demonstrating strong physics and reasoning skills. Grok 4 advanced this to around 87%, outperforming models like Claude 4.
- Simple Definition: An ultra-hard science quiz for experts—like a test for professors. Grok 3 did well, and Grok 4 crushed it even more.
Grok (The Verb)
- Tech Definition: Coined by Robert A. Heinlein in Stranger in a Strange Land, meaning to intuitively and deeply understand something holistically—embodying Grok AI’s truth-seeking design philosophy.
- Simple Definition: A sci-fi term for “totally getting” an idea or person—like intuitively knowing a buddy’s thoughts. It’s the inspiration behind Grok’s name!
Grok 1 / 1.5 / 2 / 3 / 4
- Tech Definition: xAI’s AI model progression: Grok 1 (2023) focused on basics; 1.5 enhanced reasoning; 2 added image understanding; 3 (Feb 2025) scaled massively with multimodal and tool use; 4 (July 2025) introduced native tool integration, 130k-token context, and top benchmarks (e.g., 100% on AIME 2025 Heavy variant), surpassing o3, Claude 4, and Gemini 2.5 in math, coding, and reasoning.
- Simple Definition: Grok’s upgrade levels—like starting as a basic chatter (1), getting smarter (1.5), learning to see pictures (2), handling everything (3), and becoming a world-beater in smarts and speed (4) by mid-2025.
Grokipedia
An AI-powered online knowledge platform announced by Elon Musk in September 2025 through his company xAI. Grokipedia is envisioned as an alternative to Wikipedia, designed to provide a more accurate, neutral, and dynamically updated repository of information. Unlike Wikipedia’s community-edited model, Grokipedia aims to leverage xAI’s Grok models for automated fact-checking, real-time updates, and bias reduction. The project is positioned as part of xAI’s broader mission to accelerate scientific discovery and improve humanity’s understanding of the universe.
Hitchhiker’s Guide to the Galaxy
- Tech Definition: Douglas Adams’ satirical sci-fi novel that influenced Grok’s humorous, irreverent personality—emphasizing witty, helpful responses over bland accuracy.
- Simple Definition: A hilarious space adventure book that shaped Grok’s fun side—like teaching it to be clever and cheeky while answering questions.
HMMT (Harvard-MIT Mathematics Tournament)
- Tech Definition: A prestigious math competition. Grok 4 scored 96.7% on HMMT 2025, topping competitors like Claude (under 60%) and demonstrating superior symbolic reasoning.
- Simple Definition: A elite math battle between top schools—Grok 4 dominated it, proving it’s unbeatable in number-crunching challenges.
LCB (LiveCodeBench)
- Tech Definition: A coding evaluation benchmark. Grok 3 scored 57%, showing solid programming skills. Grok 4 improved to top rankings, second overall in coding per LMSYS tests.
- Simple Definition: A real-world coding test—like debugging puzzles. Grok 3 was good, but Grok 4 became a coding champ.
Macrohard
A tongue-in-cheek named initiative by Elon Musk under xAI, described as a “purely AI software company” aimed at simulating entire software firms like Microsoft using artificial intelligence alone, without physical hardware production. It evolved from a 2021 meme tweet (“Macrohard >> Microsoft”) into an active project in 2025, with ongoing hiring for AI-driven software development and positioning as a stronger AI equivalent to traditional tech giants.
Memory Feature
- Tech Definition: Grok’s conversation retention system that stores prior interactions for context-aware responses, manageable via settings for privacy (e.g., clear history with the book icon).
- Simple Definition: Grok’s built-in recall—like remembering your last chat so it can pick up where you left off, but you can wipe it for fresh starts.
Multimodal
- Tech Definition: xAI’s paid subscription on grok.com, offering higher usage quotas beyond free limits. Pricing and details at https://x.ai/grok (distinct from X Premium+).
- Simple Definition: Grok juggling different stuff at once—like reading words, viewing pics, or hearing voices—to give fuller answers.
SuperGrok
- Tech Definition: xAI’s paid subscription on grok.com, offering higher usage quotas beyond free limits. Pricing and details at https://x.ai/grok (distinct from X Premium+).
- Simple Definition: A premium plan for more Grok chats—like an upgrade pass for unlimited AI time without hitting caps.
Think Mode
- Tech Definition: A feature in Grok 3+ that enables step-by-step reasoning for complex problems, simulating human deliberation—enhanced in Grok 4 for faster, more accurate logic in math or puzzles.
- Simple Definition: Grok’s thoughtful pause button—like breaking down tough questions one step at a time for better results.
USAMO (U.S.A. Mathematical Olympiad)
- Tech Definition: An advanced math olympiad benchmark. Grok 4 scored nearly triple GPT-4’s on USAMO 2025, emphasizing visual and symbolic reasoning improvements.
- Simple Definition: A national math genius contest—Grok 4 blew past older AIs, showing off its puzzle-solving prowess.
Voice Mode
- Tech Definition: Grok’s audio interaction feature, available on iOS and Android apps, allowing spoken queries and responses with natural intonation—part of its multimodal evolution.
- Simple Definition: Talking to Grok like a friend—speak your question on the app, and it talks back, hands-free!
X Premium+
- Tech Definition: X’s (formerly Twitter) top-tier subscription providing early and expanded access to Grok features, tied to the platform’s ecosystem for seamless integration.
- Simple Definition: X’s VIP membership—pay monthly for priority Grok perks, like chatting more and getting updates first.
xAI
- Tech Definition: Founded in 2023 by Elon Musk to accelerate scientific discovery, xAI builds Grok models with a focus on truth-seeking and integration with X for real-time data.
- Simple Definition: The innovative team behind Grok—like a squad of brainiacs pushing AI to help humanity uncover big truths.
