- The Summary AI
- Posts
- 🔥 xAI’s Grok 3 Breaks Records
🔥 xAI’s Grok 3 Breaks Records
PLUS: Anthropic Prepares Biggest AI Update

Welcome back!
xAI’s Grok 3 just set a new benchmark in AI performance, surpassing all other models in public leaderboards. With a 200,000-GPU training powerhouse and upcoming voice capabilities, xAI is proving that small, focused teams can obtain state-of-the-art results, as DeepSeek also recently demonstrated. This could accelerate the upcoming releases from OpenAI, Google, and Anthropic. Let’s unpack…
Today’s Summary:
🚀 xAI's Grok 3 dominates AI rankings
📚 Perplexity launches free research tool
💡 Anthropic teases its most advanced AI model
📝 LLaDA: First diffusion model for text generation
📽️ Step-Video-T2V: New open-source video AI
📂 OpenAI adds file & image uploads to ChatGPT
🛠️ 2 new tools

TOP STORY
xAI's Grok 3 Tops AI Performance Charts
The Summary: xAI has released Grok 3, marking a major leap in AI model performance. Trained using a massive 200,000 GPU data center in Memphis, the new model family includes specialized versions for quick responses and reasoning. Early tests show Grok 3 securing the top spot in public benchmarks.
Key details:
Grok 3 topped the Arena leaderboard with a 1400+ score, becoming the first model to reach this milestone
Independent testing by Andrej Karpathy confirms strong performance in complex tasks like board game logic and mathematical calculations
Voice mode coming next week, API access in the coming weeks
Available for X Premium+ subscribers ($22/month), with SuperGrok tier ($30/month) unlocking advanced features
xAI plans to open-source previous Grok 2 model once Grok 3 stabilizes
Why it matters: The speed at which xAI reached state-of-the-art performance - just one year after starting from scratch - sets a new precedent in AI development. This acceleration is likely to push OpenAI, Google DeepMind, and Anthropic to ramp up their upcoming releases, driving even more ambitious advancements in the field.

PERPLEXITY AI
Perplexity challenges OpenAI with free research tool
The Summary: Perplexity has launched Deep Research, an AI tool that conducts deep research, scanning hundreds of sources to generate reports in minutes. Unlike competitors OpenAI and Google, Perplexity offers free access to all users, with unlimited queries for Pro subscribers at $20/month. The tool dynamically searches, reads, and refines its research approach.
Key details:
Free users get limited daily queries, while Pro subscribers ($20/month) get 500 queries daily - compared to OpenAI's $200/month and Google's $19.99/month
Scores 21.1% on Humanity's Last Exam benchmark, second only to OpenAI's 26.6%, and ahead of Gemini 7.2%
Completes most research tasks in under 3 minutes vs 5-30 minutes for OpenAI's Deep Research
Uses DeepSeek-R1 model to achieve cost efficiency, enabling 10-100x lower pricing than competitors
Why it matters: Perplexity is positioning itself as an accessible alternative to costly AI research tools, offering fast, in-depth analysis at a fraction of the price.

ANTHROPIC
Anthropic's biggest model update nears release
The Summary: Anthropic plans to launch its new AI model soon, combining quick responses and complex reasoning in a single system. The model will feature an adjustable control, allowing developers to balance speed, cost, and processing depth. Early tests show it beats OpenAI's o3-mini-high.
Key details:
Model uses a "sliding scale" to control computing costs and reasoning
Amodei draws a parallel to human cognition: "you don't have two brains”
Reduced clinical study report writing from 12 weeks to 3 days
AI progress is "a race between making models more powerful and understanding them"
Why it matters: This approach could end the artificial divide between "chat" AI models and "reasoning" AI models. The sliding scale feature gives developers control over AI resource allocation while maintaining consistent model behavior.
“Possibly by 2026 or 2027, the capabilities of AI systems will be best thought of as akin to an entirely new state populated by highly intelligent people appearing on the global stage—a country of geniuses in a datacenter"

QUICK NEWS
Quick news
LLaDA is the first diffusion model able to generate text
Step-Video-T2V is a state-of-the-art open source video AI
OpenAI o1 and o3-mini now supports file & image uploads in ChatGPT

TOOLS
🥇 New tools

That’s all for today!
If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/