Welcome back! A code editor built its own frontier-class model for a tenth of the price, Google turned its testing playground into a full development platform overnight, Jeff Bezos is reportedly raising the largest private fund in history to buy and automate physical industries and Anthropic ran the biggest qualitative AI study ever conducted using Claude as the interviewer.
In today’s Generative AI Newsletter:
Cursor: Can a code editor beat the frontier labs at their own game?
Google: Did AI Studio become the best app builder?
Bezos: What do you buy with $100B and a fleet of AI agents?
Anthropic: What do 81,000 people actually think about AI?
Latest Developments
Cursor Launches Composer 2 and Takes On the Frontier Labs
Anysphere, the company behind AI code editor Cursor, shipped Composer 2 this week. It is a third-generation in-house model trained exclusively on code that now sits within striking distance of the most expensive frontier models at a fraction of the cost.
Benchmark Performance: Composer 2 scored 61.7% on the independent Terminal-Bench 2.0, beating Anthropic's Opus 4.6 at 58% and landing within five points of GPT-5.4 on Cursor's own internal benchmark.
Pricing: At $0.50 per million input tokens and $2.50 per million output tokens, Composer 2 costs roughly a tenth of GPT-5.4 and a twentieth of Opus 4.6 at comparable speeds.
Speed: The model runs 2x faster than its predecessor while maintaining or improving quality across vision and tool-calling tasks.
Progression: Scores on Cursor's internal CursorBench have climbed from 38% to 61.3% across three model generations shipped since October.
This is a significant moment. An application-layer company building a model that competes with dedicated frontier labs on coding tasks changes the economics for every developer currently paying full price for Opus or GPT-5.4. If the trend holds, the best coding model may not come from a model lab at all.
Special highlight from our network
That's vector search. It turns data into numbers that represent meaning, so "rainy weather shoes" finds "waterproof boots" automatically. It's the hidden engine behind every smart AI app being built right now.
MongoDB's AI & Vector Search course teaches you how to build this yourself, from scratch. If you dream of building something artificially intelligent, this is one of those key skills that unlock everything.
Special highlight from our network
For the last ~3 years, the AI industry has been racing to build bigger, smarter models like ChatGPT, Claude, Gemini and so on. The assumption was simple: whoever built the most powerful model wins.
But that race is actually already slowing down.
Now the industry is beginning to look beyond the model race and ask: how do companies turn AI into something that actually works and brings value?
In this LinkedIn Live, Ori Goshen (Co-Founder & Co-CEO of AI21 Labs) joins Steve Nouri to unpack what the Post-Model Era means for enterprise AI, from moving beyond the model race to building systems that are reliable, practical and ready for production.
Wondering where AI is heading next? This conversation will help put the direction into perspective.
Join LinkedIn Live to find the answers.
Google Turns AI Studio Into a Full-Stack App Builder
Google rebuilt AI Studio from the ground up this week, transforming what was previously a prompt testing playground into a complete development environment powered by its Antigravity coding agent.
Full Stack: The new version pairs Antigravity with Firebase, meaning the AI automatically detects when an app needs user accounts or data storage and configures the backend for you.
Prompt to Production: Users can describe an app in plain English and go from prompt to a deployed, live application without touching infrastructure.
Roadmap: Google previewed upcoming integrations with Workspace (Drive, Sheets), one-click deployment and connections to payment processors and live data sources.
Desktop Expansion: Google is also testing a dedicated Gemini desktop app for Mac, targeting ChatGPT and Claude's desktop presence directly.
Internal teams have reportedly already built hundreds of thousands of apps on the platform. The timing is notable. On the same day, OpenAI revealed plans to merge ChatGPT, Codex and Atlas into a single desktop superapp. Both companies appear to have reached the same conclusion: the era of separate AI tools is ending.
Bezos Reportedly Raising $100B for Project Prometheus
Jeff Bezos is reportedly raising a $100 billion fund to buy manufacturers in aerospace, chipmaking and defence, then automate their operations with AI. He has reportedly travelled to the Middle East and Singapore to fundraise.
Scale: If completed, this would be the largest private investment fund in history.
Target Sectors: The acquisitions would focus on physical industries including chip fabrication, defence contractors and aerospace manufacturers.
Strategy: The plan centres on buying companies with existing manufacturing capacity and retrofitting their operations with AI-driven automation.
Secrecy: Details remain thin. The project is reportedly being run outside of Amazon under the working name Project Prometheus.
The sheer scale of the fund signals a bet that AI's biggest returns will not come from software alone. Bezos appears to be positioning for a world where AI-driven automation of physical production becomes as valuable as the models themselves.
Anthropic Runs the Largest Qualitative AI Survey Ever Conducted
Anthropic released what it says is the biggest qualitative study on AI attitudes ever produced. The company used a specially built version of Claude to interview 81,000 of its own users across 159 countries and 70 languages in a single week.
Top Hope: Professional excellence was the most frequently cited aspiration, followed by freeing up time, financial independence and broader life management.
Top Fear: AI getting things wrong outranked every other concern. Job anxiety, losing personal agency and over-reliance followed closely.
Regional Splits: India and South America skewed above average in optimism. The US, Europe, Japan and South Korea ran neutral or below.
Methodology: Claude conducted open-ended conversations rather than fixed surveys, producing qualitative depth at a scale that would be impossible with human interviewers.
The headline finding is that most people are not choosing between hope and fear. They are carrying both at the same time. But almost as notable is the proof of concept: Claude running 80,000 in-depth interviews across 70 languages in seven days is a research capability that simply did not exist a year ago.
TOP 5 Tools of the Week
Vidu: an all-in-one AI image and video creation platform that turns ideas into fast, high-quality visuals with tools for text-to-video, image-to-video, and consistent character-driven storytelling.
Anam AI: builds real-time AI Personas that make digital interactions feel human. Intuitive, expressive, and built for conversations that finally feel natural.
Apify: is a full-stack web scraping and data extraction platform that helps teams get real-time web data fast for AI apps, research, lead generation, and automation.
Coursebox AI: is an AI course creation platform that turns documents, videos, and links into interactive, branded training courses in minutes.
NetMind AI: is a unified AI platform that helps teams build, deploy, and scale real AI fast with flexible infrastructure, model serving, and enterprise-ready tools in one place.
