TLDR AI 2025-11-10

ChatGPT is only one part of OpenAI's success. The other part is the breathtaking speed with which OpenAI built a new monetization model.

In a fireside chat at Monetize 2025, OpenAI's Head of Financial Engineering, Sara Conlon, explained how she built an engineering billing org poised for hypergrowth.

Read the blog to learn:

• Why centralizated monetization infrastructure is essential
• Hard lessons from scaling fast, including outages and high-stakes incidents
• How OpenAI structures its billing team
• Preparing for the future: hybrid models, iteration velocity, and systems built for coexistence

Read The OpenAI Monetization Blueprint on the Metronome blog

Nano Banana 2 is expected to launch on November 11. The model will have 2K native output and huge improvements across many areas. This article provides some examples of outputs from the new model along with comparisons with outputs from Nano Banana 1.

GPT-5-Codex-Mini, a more compact and cost-efficient version of GPT-5-Codex that allows roughly 4x more usage at a slight capability tradeoff, is now available in OpenAI's CLI and IDE extension. OpenAI has increased rate limits by 50% for ChatGPT Plus, Business, and Edu users. There is now priority processing for ChatGPT Pro and Enterprise customers.

OpenAI has committed to spending $1.15 trillion on hardware and cloud infrastructure between 2025 and 2035. The estimated annual spending and commitments convey an absolutely enormous level of potential and ambition. OpenAI will need to grow from around $10 billion in 2024 revenue to $577 billion by 2029. This article takes a look at the numbers.

Most organizations are still in the experimentation or piloting phase. High performers use AI to drive growth, innovation, and cost. Redesigning workflows is a key success factor. The potential impact of AI on employment is still debated - 43% of workers in a survey expected no change in the overall size of the workforce in their organizations in the coming year.

IT complexity can drive costs up across your company. Intelligent IT automation is key to taming technology chaos. Get the insights you need from the IBM Institute for Business Value. 👉 Read the IBV report

Continual learning, the ability to acquire new knowledge without forgetting old information, remains a major obstacle to AGI since current models suffer from “catastrophic forgetting,” where learning new tasks diminishes proficiency on previous ones. A proof-of-concept architecture called Hope, a self-modifying model that optimizes its own memory, outperformed modern recurrent models and transformers on language modeling and excelled particularly at long-context needle-in-a-haystack tasks.

This repository contains a curated reference library of production-tested Claude Code infrastructure. It contains production-tested infrastructure for auto-activating skills via locks, specialized agents for complex tasks, and more. Using this infrastructure, skills will suggest themselves based on context, hooks will trigger skills at the right time, modular skills stay under context limits, dev docs preserve knowledge across resets, and agents will streamline complex tasks.

K2-Thinking's release has led many developers to be curious about its native INT4 quantization format. INT4 accelerates RL training due to its low-latency profile. Quantization is no longer a tradeoff. Native low-bit quantization will become a standard paradigm for large model training with the evolution of param-scaling and test-time-scaling.

Companies are pouring billions of dollars into cutting-edge computing facilities. The sheer scale of spending is causing investors to be concerned. Tech giants are turning to financing maneuvers that may add to the risk. Hyperscalers are leveraging a growing list of complex debt-financing options, including corporate set, securitization markets, private financing, and off-balance-sheet vehicles.

The gap between everyday use of AI and its core capabilities is massive. Chatbots are used to write emails or replace web searches, while the underlying models outperform human experts in math and coding competitions. OpenAI is advocating for light regulation now to cement its market position before the next leap in capability. Once systems become transformative enough to warrant a coordinated national response, there will be a shift to federal oversight that overrides state-level restrictions.

Learn how to: enforce fine-grained access control to ensure on AI agents, enable agents to call APIs on users' behalf with securely managed tokens, and more. Watch the Auth0 technical demo

A letter from OpenAI's policy chief to the Trump administration proposes expanding the Chips Act beyond semiconductors to “de-risk” data-center buildouts and help close the “electron gap” with China, which added 429 GW of power capacity in 2024 compared to America's 51 GW.

Hybrid LLMs that combine attention with alternatives such as Mamba or linear attention are now first-class citizens in vLLM V1, enabling efficient long-context inference for models including Qwen3-Next and Granite 4.0.

The releases aim to address long-standing pain points in testing and optimizing AI agents.

Investment into new subsea cable projects is expected to reach around $13 billion between 2025 and 2027.