21 July 2025
Topic: Models And Releases
ChatGPT users send 2.5 billion prompts a day
ChatGPT receives 2.5 billion prompts from global users every day....
TL;DR
ChatGPT has seen significantly rapid growth, with daily query counts more than doubling to 2.5 billion in around 8 months, surpassing Google's daily search count estimates.
Key Takeaways:
Key Takeaways:
- ChatGPT receives 2.5 billion prompts globally every day, with 330 million coming from the US.
- Google receives an estimated 13.7-16.4 billion daily searches, with only 14 billion being the most conservative estimate from Google itself.
- ChatGPT's daily query count has more than doubled in approximately 8 months, with initial numbers reaching over 1 billion in December.
A ChatGPT ‘router’ that automatically selects the right OpenAI model for your job appears imminent
Like going to the supermarket and staring at aisles of cereal and sauces, the average ChatGPT user is currently faced with an overabundance....
TL;DR
OpenAI is reportedly developing a 'router' function in ChatGPT to automatically select the best AI model for user input, alleviating the 'overabundance of choice' problem for users.
Key Takeaways:
Key Takeaways:
- The new 'router' function will automatically select the best OpenAI model to respond to a user's input on the fly.
- This feature will enable users to receive smarter, more intelligent, or more capable responses from ChatGPT without needing to manually select a model.
- The introduction of the 'router' function may increase public adoption and acceptance of AI by simplifying the user experience and highlighting the benefits of using more advanced models.
Show HN: Lotas – Cursor for RStudio
Hey HN! We’re Jorge and Will from Lotas (https://www.lotas.ai/), and we’ve built an AI coding assistant into RStudio (think Cursor for RStudio).RStudi...
TL;DR
Lotas introduces Rao, an AI code editor designed to accelerate RStudio workflows by providing features such as data analysis, code generation and editing, and output interpretation.
Key Takeaways:
Key Takeaways:
- Rao understands project files and data to generate targeted relevant code.
- Rao can generate, run, and edit R scripts and R markdown files.
- Rao provides comprehensive output analysis, suggesting next steps and interpreting results.
OpenAI says ChatGPT users send over 2.5 billion prompts every day
OpenAI’s ChatGPT sees more than 2.5 billion requests daily, with 330 million from users based in the US, according to data obtained by Axios. The data...
TL;DR
ChatGPT users send over 2.5 billion prompts every day, with 912.5 billion requests annually, posing a threat to Google's search dominance.
Key Takeaways:
Key Takeaways:
- ChatGPT users send over 2.5 billion daily prompts, with 912.5 billion annual requests.
- In just a few months, ChatGPT's weekly users spiked from 300 million to over 500 million.
- OpenAI's rapid growth and recent releases, such as ChatGPT Agent and a potential AI-powered web browser, pose a significant threat to Google's search engine.
LLM Alloying Improves Performance over Single Model
Article URL: https://xbow.com/blog/alloy-agents/ Comments URL: https://news.ycombinator.com/item?id=44630724 Points: 24 # Comments: 6...
TL;DR
Combining different large language models (LLMs) into an 'alloy' can significantly improve performance on complex agentic tasks, with potential applications in areas like cybersecurity.
Key Takeaways:
Key Takeaways:
- Model alloys can improve performance by combining the strengths of different models, particularly when the models are diverse and have low correlation in solve rates.
- Alloys tend to outperform individual models, especially when the better individual model is balanced within the alloy.
- Model alloys should be considered for tasks that require a combination of different ideas and have at least a double-digit number of model calls, but may not be suitable for tasks with steady progress or single-model superiority.
Grok’s AI companions drove downloads, but its latest model is the one making money
Given Grok's expensive new subscription offering, timed alongside the Grok 4 launch, it's not surprising to see that even a smaller increase in the nu...
TL;DR
xAI's latest AI model, Grok 4, has driven a 325% increase in iOS revenue to $419,000 within three days of its launch, and 279% increase in daily downloads to 197,000.
Key Takeaways:
Key Takeaways:
- Grok 4's launch led to a 325% increase in iOS revenue to $419,000 within three days.
- Daily downloads of the Grok iOS app increased 279% to 197,000.
- The addition of Grok's AI companions led to a smaller increase in downloads (40%) and revenue (9%) compared to the launch of Grok 4.
Chrome will let you switch between personal and work accounts on iOS
Google is introducing the option to easily switch between personal and work profiles in its Chrome browser on iOS, with data kept separate between the...
TL;DR
Google introduces the option to switch between personal and work profiles in its Chrome browser on iOS, separated by data with enterprise accounts.
Key Takeaways:
Key Takeaways:
- The feature is only available to users with managed enterprise accounts.
- Browsing data, including tabs, history, and passwords, remains separate between personal and work profiles.
- This feature is not available on Android, where a device-wide work profile supports Google apps, including Chrome.
Chrome for iOS makes it easier to switch between work and personal Google accounts
When users sign in or switch to their work Google account in Chrome, browsing related to their organization will be kept separate from their other bro...
TL;DR
Google introduces a new feature for iOS users to easily switch between work and personal Google accounts in Chrome, improving account separation and security.
Key Takeaways:
Key Takeaways:
- The new feature provides strict data separation between work and personal browsing, keeping local data and content within the managed account browsing experience.
- Google has incorporated advanced security features, including URL filtering on iOS, that allow organizations to block prohibited GenAI sites and redirect employees to approved corporate services.
- Chrome Enterprise's reporting capabilities are now extended to both Android and iOS, enabling organizations to send security event data to their preferred security tools.
[New Architecture] Hierarchical Reasoning Model
Rockchip unveils RK182X LLM co-processor: Runs Qwen 2.5 7B at 50TPS decode, 800TPS prompt processing
[D] Gemini officially achieves gold-medal standard at the International Mathematical Olympiad
NVIDIA Brings Reasoning Models to Consumers Ranging from 1.5B to 32B Parameters
Google Had second system score gold without access to training corpus or hints, just pure natural language
Gemini Deep Think achieved Gold at IMO
Kimi K2 is already irrelevant, and it's only been like 1 week. Qwen has updated Qwen-3-235B, and it outperforms K2 at less than 1/4th the size
Gemini with Deep Think achieves gold medal-level
Imminent release from Qwen tonight
So the consensus right now is: Gemini better for coding, ChatGPT better for enquiries, information and philosophy?