Why Xai Is Giving You 'limited' Free Access To Grok 4

ZDNET

ZDNET's cardinal takeaways:

• Grok 4 is now accessible by default for unpaid xAI users
• Restrictions astir nan "limited" free entree play are unclear.
• The move appears to beryllium an effort to pull much paying users.

Elon Musk's AI start-up xAI has made its latest model, Grok 4, disposable to each users -- some paid and unpaid. 

Grok now automatically opens successful Auto mode for users of nan chatbot's free tier, meaning nan strategy automatically determines whether aliases not a fixed query needs Grok 4's precocious reasoning capabilities, aliases if a much constricted exemplary will beryllium up to nan occupation (a akin attack to what OpenAI's caller GPT-5 model aims to do). Users tin besides manually prime "Fast" mode, which generates responses much quickly utilizing Grok 3, aliases "Expert" mode, which exclusively uses Grok 4, taking much clip to travel up pinch higher-quality responses.

Also: GPT-5 bombed my coding tests, but redeemed itself pinch codification analysis

The institution announced nan "limited" merchandise successful an X post connected Sunday, though it didn't specify nan number of messages that unpaid users could speech pinch nan chatbot complete a fixed clip play aliases immoderate different restrictions.

"For a constricted time, we are rolling retired generous usage limits truthful you tin research Grok 4's afloat potential," nan station read.

Even Grok 4 itself was uncertain astir nan rules. "Exact limits for Grok-4's free tier are unclear, arsenic xAI hasn't publically elaborate them, and they whitethorn alteration by region aliases relationship type," nan chatbot wrote erstwhile I asked astir nan limits of nan free entree period. "To beryllium safe, expect astir 5 to 10 messages per 12 hours pinch Grok-4 connected nan free tier, but you mightiness deed nan limit sooner if you usage analyzable queries aliases during highest times."

When xAI unveiled Grok 4 past month, Musk described it arsenic "the world's astir powerful AI model," claiming it outperformed frontier models for illustration Google's Gemini 2.5 Pro and OpenAI's 4o connected respective cardinal benchmarks, including Humanity's Last Exam.

Also: How Google's Genie 3 could alteration AI video - and fto you build your ain interactive worlds

Subscriptions to Grok's subscription tiers, SuperGrok and SuperGrok Heavy, still costs $30 per period and $300 per month, respectively.

Covering nan precocious costs of AI

By temporarily making Grok 4 disposable to each users, xAI appears to beryllium angling successful portion to get its caller exemplary into nan hands of arsenic galore group arsenic imaginable -- a maneuver ZDNET elder editor Sabrina Ortiz has been tracking. A percent of those group whitethorn beryllium impressed capable by nan caller model's abilities that, erstwhile they deed nan limits of nan free entree window, they'll determine that it's worthy upgrading to a paid subscription. 

xAI besides plans to present ads into Grok's responses to thief salary for nan precocious costs of moving and refining nan model, nan Financial Times reported Thursday.

Also: You tin effort Gemini Live successful your favourite Google apps now, and it blew maine away

Musk announced successful different X post Sunday that xAI users tin now "long press" -- holding down connected an icon for longer than conscionable nan accustomed speedy pat of nan thumb -- connected immoderate image to propulsion up an action to move it into a video utilizing AI. This could besides yet beryllium a play to tie successful advertisers: Amazon precocious debuted a akin AI-powered text-to-video feature. 

Grok 4 vs. GPT-5

The impermanent merchandise of Grok 4 to unpaid users comes conscionable days aft OpenAI -- pinch whom Musk has agelong had a very nationalist squabble -- released GPT-5 to each users astatine once, a caller attack for nan company. 

Also: GPT-5 bombed my coding tests, but redeemed itself pinch codification analysis

As of Monday, GPT-5 occupied nan apical spot successful aggregate captious categories (including text-generation, vision, and coding) connected LMArena, an online level wherever users tin complaint and comparison nan capacity of ample connection models. OpenAI's o3 exemplary besides defeated Grok 4 successful nan last information of Google Game Arena's chess tourney past week.

Grok 4, however, outperformed GPT-5 by a hairsbreadth successful nan ARC-AGI-2 -- a trial designed to gauge AI models' high-level reasoning capabilities -- though astatine a overmuch higher costs ($2.17 per task vs. $0.73 per task).