OpenAI’s GPT-4.5 Debuts—With a GPU Bottleneck

OpenAI has launched GPT-4.5, its most advanced large language model yet—but access is limited due to a GPU shortage. The model is rolling out first to ChatGPT Pro subscribers ($200/month) and API developers, with ChatGPT Plus users expected to gain access next week.

The big picture:
Massive computational demands: GPT-4.5 is OpenAI’s most expensive model yet, costing $75 per million input tokens and $150 per million output tokens—far pricier than GPT-4o. CEO Sam Altman admitted OpenAI simply lacks enough GPUs to scale it further right now.

Not a reasoning model: Unlike new AI models from OpenAI’s rivals, GPT-4.5 does not feature chain-of-thought reasoning but instead refines traditional LLM capabilities, focusing on natural conversation and creative responses.

Improved efficiency: OpenAI claims GPT-4.5 is 10x more computationally efficient than GPT-4, reducing hallucination rates significantly (37.1% vs. GPT-4o’s 61.8%).

The debate:
Early reviews are mixed. Some researchers argue GPT-4.5 delivers only marginal benchmark improvements over existing models, while others highlight its qualitative advancements in fluency, creativity, and human-like interaction. A leaked system card originally stated GPT-4.5 is “not a frontier model,” raising questions about its true positioning in OpenAI’s AI lineup.

Why it matters:
With competitors like Anthropic, Google, and DeepSeek launching powerful models, OpenAI faces mounting pressure to stay ahead. The GPU shortage underscores the industry-wide infrastructure challenge, which OpenAI hopes to address through its ambitious Stargate data center project.

The bottom line:
GPT-4.5 is a step forward, but it’s not a revolution. While it offers improvements in usability and efficiency, questions remain about whether its high cost justifies its incremental progress.

Meta to Launch Standalone AI App

Meta is preparing to launch a dedicated Meta AI app, placing its chatbot alongside Facebook, Instagram, and WhatsApp in its ecosystem. The move, expected in Q2, underscores Mark Zuckerberg’s ambition to position Meta AI as the most widely used digital assistant—potentially surpassing ChatGPT and Google’s Gemini.

THE BIG PICTURE:
Meta AI’s rapid growth – Since launching in 2023, Meta AI has been integrated into Facebook, Instagram, WhatsApp, and Messenger, replacing search functions to keep users engaged. It now has 700 million monthly active users, with India as its largest market.

A push for deeper engagement – Unlike rivals, Meta AI has lacked a standalone app, making direct comparisons to ChatGPT difficult. Analysts say its website traffic remains low compared to major competitors. A dedicated app could centralize conversations, improve personalization, and expand its reach across devices like Meta’s Ray-Ban smart glasses.

Monetization on the horizon – Meta is exploring a paid subscription model for Meta AI, similar to OpenAI and Microsoft’s premium chatbot tiers. CFO Susan Li has hinted at future revenue streams, including paid recommendations and premium AI features.

THE BROADER CONTEXT:
Tech giants are racing to dominate the AI assistant space. Google and Elon Musk’s xAI have already rolled out individual apps for Gemini and Grok. Meanwhile, ChatGPT remains the most downloaded AI app.

THE BOTTOM LINE:
Zuckerberg is betting big on Meta AI, pushing his teams to work at breakneck speed to make it the dominant chatbot by year’s end. With a standalone app and monetization plans in the works, Meta AI is gearing up for an aggressive expansion—but it still has ground to cover to match its biggest rivals.

OpenAI Plans to Bring Sora Video Generation to ChatGPT

OpenAI is preparing to integrate its AI video tool, Sora, into ChatGPT, aiming to make AI-powered video creation more accessible to everyday users. While no timeline has been set, OpenAI’s product lead Rohan Sahai confirmed during a Discord office hours session that Sora’s capabilities will eventually be embedded into the chatbot.

THE BIG PICTURE:
Expanding Sora’s reach – Currently available only as a standalone web app, Sora generates up to 20-second cinematic clips. Initially marketed to video professionals, OpenAI is now broadening its appeal by positioning it within ChatGPT.

Strategic move for ChatGPT – Embedding Sora into ChatGPT could attract more users and encourage upgrades to premium subscription tiers, which may offer higher video generation limits.

Balancing simplicity and functionality – OpenAI initially kept Sora separate to maintain ChatGPT’s streamlined experience. The ChatGPT version may have fewer editing tools than the standalone web app.

WHAT’S NEXT:
A mobile app on the horizon – OpenAI is exploring a dedicated mobile app for Sora, actively hiring engineers for the project.

Beyond video: AI-powered images – OpenAI is developing an image generator powered by Sora, potentially offering more photorealistic outputs than DALL·E 3.

Sora Turbo upgrade – OpenAI is working on a new version of Sora’s model, hinting at improved speed and quality.

THE BOTTOM LINE:
By integrating Sora into ChatGPT, OpenAI is making a play to merge text and video AI into a single, consumer-friendly experience. While details remain scarce, the move underscores OpenAI’s push to redefine how users interact with generative media.

Murati’s New AI Lab

Google is integrating its AI model, Gemini, into Sheets, bringing advanced data analysis and visualization tools directly into spreadsheets. The upgrade, now available to all Workspace business users, aims to simplify complex data tasks with AI-powered insights.

THE BIG PICTURE:
Instant insights: Gemini can analyze data trends, detect anomalies, and generate correlations, helping users make sense of their spreadsheets faster.

AI-generated charts: Users can create visualizations like heatmaps and insert them as static images in their spreadsheets.

Conversational queries: By clicking the Gemini icon, users can ask for predictions, trend analyses, or performance breakdowns—like forecasting net income or identifying top-performing marketing channels.

HOW IT WORKS:
For advanced tasks: Gemini runs Python code in the background, executing multi-layered analyses without requiring users to write code themselves.

For simpler requests: The AI can still rely on traditional spreadsheet formulas to generate insights.

Data hygiene matters: To ensure accuracy, Google advises keeping spreadsheets well-structured with consistent formats and clear headers.

THE BOTTOM LINE:
By embedding Gemini into Sheets, Google is turning spreadsheets into an AI-powered analytics tool. Whether this makes business users more efficient—or just more reliant on AI—remains to be seen.