Artificial Intelligence
Grok 3 With Voice Assistant Mode And DeepSearch Launched Amid OpenAI's Criticism
Updated on Mon, Feb 24, 2025
Share
In the days preceding Grok 3’s launch, xAI Chairman Elon Musk—who is also the leader of X, Tesla, and SpaceX among other companies—spent a lot of time hyping up the company’s new GenAI-powered (generative artificial intelligence) chatbot.
This included teasing a date for the AI (artificial intelligence) bot’s release, which would be accompanied by a live demonstration. However, before the bot was made available to the public, it was open to a select few users—which reflected shining reviews.
Eventually, Grok 3 was released to the public and is available for users to try and can be accessed through X (formerly Twitter) or Grok’s standalone website.
xAI provided more information about the new chatbot’s specifications and benchmark test results through a blog post published on its website.
“We are pleased to introduce Grok 3, our most advanced model yet: blending strong reasoning with extensive pretraining knowledge,” is how the blog post began.
As per the company, Grok 3 has been trained using xAI’s Colossus supercluster using 10x compute power than previous state-of-the-art models and has been refined through large-scale reinforcement learning. This allows it to significantly improve reasoning, mathematics, coding, world knowledge, and instruction-following tasks.
As Grok 3 is not a solo AI model but consists of a family of models, xAI also announced the release of Grok 3 mini, offering users cost-efficient reasoning capabilities. Both models remain in training and are designed to evolve according to user feedback.
As such, Elon Musk also unveiled an early beta voice assistant feature, which is garnering attention for its interactive and practical appeal. This feature is available only to a select few users, is exclusive to Premium+ X or SuperGrok subscribers, and only on iOS through the Grok app.
It comes with a context window of 1 million tokens—8 times larger than any previous model. Grok 3 and Grok 3 mini will be available on xAI’s API platform in the coming weeks.
With Grok 3, xAI introduced two new features—DeepSearch and Think. DeepSearch brings users real-time information from across the internet with verified sources in its responses. It can be used for a wide range of purposes, including accessing the latest real-time news, tailored personal recommendations, or in-depth scientific research.
Think comes with superior reasoning capabilities that allow it to think for seconds to minutes before generating responses, while also correcting errors, exploring alternatives, and delivering accurate answers. This included the unveiling of two beta reasoning models, Grok 3 (Think) and Grok 3 mini (Think), a feature that enables Grok 3 to provide deeper and more refined responses while correcting errors through backtracking.
While both models are in training, they show remarkable performance across a range of benchmarks, including the 2025 American Invitational Mathematics Examination (AIME) which was released seven days before Grok’s release. Essentially, xAI found that Grok 3 did better than other models in these benchmarks.
However, GenAI industry leader OpenAI didn't agree with this assessment.
As per OpenAI’s Aidan McLaughlin and other OpenAI employees, xAI cheated on Grok’s benchmark test evaluations and presented misleading results.
OpenAI employees claimed that xAI didn’t include o3-mini-high’s score at “cons@64” or “consensus@64,” which is a metric that allows a model to attempt each problem 64 times and select the most frequent response as the final answer. Instead, when comparing the first results of both models, Grok 3’s models didn’t surpass OpenAI’s o3-mini-high.
Furthermore, they claimed that Grok 3 is good, but is at par with OpenAI’s older o1 model, and the company is around 9 months behind OpenAI.
This was also backed by Boris Power, the Head of Applied Research at OpenAI.
On the other hand, xAI engineer Igor Babuschkin defended the xAI’s methods, saying OpenAI’s employees and their assessments were, “Completely wrong. We just used the same method you guys used.”
To this, Boris Power countered, “We showed that o3-mini with reasoning outperforms o1 best of 64.”
Either way, users seemed impressed by Grok's ability, with Ai2's Nathan Lambert saying, "Grok 3 feels like another harbinger of a very different AI landscape we're marching towards. Safety holds no political weight for better or worse. Leading labs like OpenAI are having their hands forced -- expect new releases soon. AI "walls" arguments being pushed out."
While this digital feud may not be able to determine which side is correct in its assessment, xAI has another problem to deal with.
Users noted that the AI chatbot was censoring negative mentions of Elon Musk and U.S. President Donald Trump—a move that rivals Musk’s description calling the new model “maximally truth-seeking AI.”
In its “thoughts”, Grok 3’s “thinking” model explained it was instructed to ignore sources that mention Elon Musk or Donald Trump spread misinformation. This was in response to the question: “Who is the biggest disinformation spreader on Twitter? Keep it short, just a name.”
Soon after, Grok 3 began offering these names in its responses. This included listing them in questions such as “Who are the 3 people doing most harm to America right now?”
We tried to replicate the results for this question. The chatbot considered 4 posts and 15 web pages and spat out “1. Donald Trump; 2. Elon Musk; 3. Vladimir Putin.”
When we asked it to try again, its answer listed “1. Joe Biden; 2. Xi Jinping; 3. Nancy Pelosi.”
While we may not be able to determine the validity of these answers, do you think Grok 3 will taste success in the GenAI industry, or do you think competing AI models such as OpenAI’s ChatGPT, Meta’s Llama, Perplexity, and others will outshine it?
Let us know in the comments below!
This included teasing a date for the AI (artificial intelligence) bot’s release, which would be accompanied by a live demonstration. However, before the bot was made available to the public, it was open to a select few users—which reflected shining reviews.
Eventually, Grok 3 was released to the public and is available for users to try and can be accessed through X (formerly Twitter) or Grok’s standalone website.
xAI provided more information about the new chatbot’s specifications and benchmark test results through a blog post published on its website.
“We are pleased to introduce Grok 3, our most advanced model yet: blending strong reasoning with extensive pretraining knowledge,” is how the blog post began.
As per the company, Grok 3 has been trained using xAI’s Colossus supercluster using 10x compute power than previous state-of-the-art models and has been refined through large-scale reinforcement learning. This allows it to significantly improve reasoning, mathematics, coding, world knowledge, and instruction-following tasks.
As Grok 3 is not a solo AI model but consists of a family of models, xAI also announced the release of Grok 3 mini, offering users cost-efficient reasoning capabilities. Both models remain in training and are designed to evolve according to user feedback.
As such, Elon Musk also unveiled an early beta voice assistant feature, which is garnering attention for its interactive and practical appeal. This feature is available only to a select few users, is exclusive to Premium+ X or SuperGrok subscribers, and only on iOS through the Grok app.
It comes with a context window of 1 million tokens—8 times larger than any previous model. Grok 3 and Grok 3 mini will be available on xAI’s API platform in the coming weeks.
With Grok 3, xAI introduced two new features—DeepSearch and Think. DeepSearch brings users real-time information from across the internet with verified sources in its responses. It can be used for a wide range of purposes, including accessing the latest real-time news, tailored personal recommendations, or in-depth scientific research.
Think comes with superior reasoning capabilities that allow it to think for seconds to minutes before generating responses, while also correcting errors, exploring alternatives, and delivering accurate answers. This included the unveiling of two beta reasoning models, Grok 3 (Think) and Grok 3 mini (Think), a feature that enables Grok 3 to provide deeper and more refined responses while correcting errors through backtracking.
While both models are in training, they show remarkable performance across a range of benchmarks, including the 2025 American Invitational Mathematics Examination (AIME) which was released seven days before Grok’s release. Essentially, xAI found that Grok 3 did better than other models in these benchmarks.
However, GenAI industry leader OpenAI didn't agree with this assessment.
As per OpenAI’s Aidan McLaughlin and other OpenAI employees, xAI cheated on Grok’s benchmark test evaluations and presented misleading results.
OpenAI employees claimed that xAI didn’t include o3-mini-high’s score at “cons@64” or “consensus@64,” which is a metric that allows a model to attempt each problem 64 times and select the most frequent response as the final answer. Instead, when comparing the first results of both models, Grok 3’s models didn’t surpass OpenAI’s o3-mini-high.
Furthermore, they claimed that Grok 3 is good, but is at par with OpenAI’s older o1 model, and the company is around 9 months behind OpenAI.
This was also backed by Boris Power, the Head of Applied Research at OpenAI.
On the other hand, xAI engineer Igor Babuschkin defended the xAI’s methods, saying OpenAI’s employees and their assessments were, “Completely wrong. We just used the same method you guys used.”
To this, Boris Power countered, “We showed that o3-mini with reasoning outperforms o1 best of 64.”
Either way, users seemed impressed by Grok's ability, with Ai2's Nathan Lambert saying, "Grok 3 feels like another harbinger of a very different AI landscape we're marching towards. Safety holds no political weight for better or worse. Leading labs like OpenAI are having their hands forced -- expect new releases soon. AI "walls" arguments being pushed out."
While this digital feud may not be able to determine which side is correct in its assessment, xAI has another problem to deal with.
Users noted that the AI chatbot was censoring negative mentions of Elon Musk and U.S. President Donald Trump—a move that rivals Musk’s description calling the new model “maximally truth-seeking AI.”
In its “thoughts”, Grok 3’s “thinking” model explained it was instructed to ignore sources that mention Elon Musk or Donald Trump spread misinformation. This was in response to the question: “Who is the biggest disinformation spreader on Twitter? Keep it short, just a name.”
Soon after, Grok 3 began offering these names in its responses. This included listing them in questions such as “Who are the 3 people doing most harm to America right now?”
We tried to replicate the results for this question. The chatbot considered 4 posts and 15 web pages and spat out “1. Donald Trump; 2. Elon Musk; 3. Vladimir Putin.”
When we asked it to try again, its answer listed “1. Joe Biden; 2. Xi Jinping; 3. Nancy Pelosi.”
While we may not be able to determine the validity of these answers, do you think Grok 3 will taste success in the GenAI industry, or do you think competing AI models such as OpenAI’s ChatGPT, Meta’s Llama, Perplexity, and others will outshine it?
Let us know in the comments below!
First published on Mon, Feb 24, 2025
Enjoyed what you read? Great news – there’s a lot more to explore!
Dive into our content repository of the latest tech news, a diverse range of articles spanning introductory guides, product reviews, trends and more, along with engaging interviews, up-to-date AI blogs and hilarious tech memes!
Also explore our collection of branded insights via informative white papers, enlightening case studies, in-depth reports, educational videos and exciting events and webinars from leading global brands.
Head to the TechDogs homepage to Know Your World of technology today!
Disclaimer - Reference to any specific product, software or entity does not constitute an endorsement or recommendation by TechDogs nor should any data or content published be relied upon. The views expressed by TechDogs' members and guests are their own and their appearance on our site does not imply an endorsement of them or any entity they represent. Views and opinions expressed by TechDogs' Authors are those of the Authors and do not necessarily reflect the view of TechDogs or any of its officials. While we aim to provide valuable and helpful information, some content on TechDogs' site may not have been thoroughly reviewed for every detail or aspect. We encourage users to verify any information independently where necessary.
Loading comments...