Search the archive

OPENAI'S NEW FUNDRAISING ROUND COULD VALUE STARTUP AT AS MUCH AS $830 BILLION

OpenAI is in the early stages of a fundraising round that could raise as much as $100 billion, valuing the startup at as much as $830 billion. It aims to complete the round by the end of the first quarter at the earliest. It is unclear whether there will be sufficient investor demand to reach the goal. The round will be one of the biggest tests OpenAI has faced since the public market's exuberance for AI spending waned.

META IS DEVELOPING NEW AI IMAGE AND VIDEO MODEL CODE-NAMED ‘MANGO'

Meta is developing a new image and video-focused AI model code-named Mango. The model is expected to be released in the first half of 2026. Image generation has proven to be a vital front in the war between the big AI model companies. It is a primary point of interest for many users and a particularly sticky feature that keeps them coming back.

INTRODUCING GPT-5.2-CODEX

OpenAI's new agentic coding model is state-of-the-art on SWE-Bench Pro and Terminal-Bench 2.0 with improved long-horizon work. OpenAI is launching a trusted access pilot to give vetted cybersecurity professionals access to future, more powerful models.

JOHN SCHULMAN ON DEAD ENDS, SCALING RL, AND BUILDING RESEARCH INSTITUTIONS (51 MINUTE VIDEO)

John Schulman estimates that with full hindsight, a few talented people could have built a ChatGPT-3.5-level model in 2018-2019 with a couple of GPU boxes. He described early OpenAI as a "rag tag" blend of small exploratory research projects and bigger engineering efforts inspired by DeepMind's AlphaGo. He expects value functions and offline RL to make a comeback, and warns that catch-up mode makes it harder to build exploratory research culture later.

AI #147: FLASH FORWARD

GPT-5.2 is a frontier model only for the frontier. OpenAI's image 1.5 looks comparable to Nano Banana Pro, and the startup has a deal for Disney's characters. Gemini 3 Flash is a good model for its speed and price. It captures the bulk of Gemini 3 Pro's intelligence quickly at a low price.

EVALUATING CHAIN-OF-THOUGHT MONITORABILITY

OpenAI has proposed a new evaluation suite to measure how reliably model reasoning can be monitored via chains-of-thought. The study assessed monitorability across 24 environments and found that reasoning transparency can vary significantly with scale, reinforcement learning, and inference-time compute.

AGENT SKILLS BECOMES AN OPEN STANDARD

Agent Skills, folders of instructions, scripts, and resources that give AI agents new capabilities on demand, originated at Anthropic (which also created MCP) and is now an open format with adoption from Cursor, GitHub, VS Code, Claude Code, and OpenAI's Codex CLI. Skills let teams package domain expertise and workflows into portable, version-controlled packages that work across different agent products.

INSIDE REPLIT'S SNAPSHOT ENGINE: THE TECH MAKING AI AGENTS SAFE

Replit built a compute and storage fabric that allows it to make changes in an isolated, reversible way. These primitives enable developers to experiment more frequently and faster. The company realized the same primitives could be used to superpower coding agents when it built Replit Agent in 2024. The system helps the human driving the agent, and the agent itself greatly benefits from the tools. This post explores the underlying systems that make the Replit Agent safe and how Replit uses them. It also takes a peek at Replit's near-term roadmap.

INTRODUCING MISTRAL OCR 3

Mistral OCR 3 is designed to extract text and embedded images from a wide range of documents with exceptional fidelity. It is a major upgrade over the previous version in forms, handwritten content, low-quality scans, and tables. Mistral OCR 3 enables downstream systems to understand structure as well as document content. The model can be integrated via API and used with Document AI, a UI that parses documents into text or structured JSON.

RUBRICS AS REWARDS: REINFORCEMENT LEARNING BEYOND VERIFIABLE DOMAINS

Scale AI researchers developed a structured approach to reinforcement learning that uses checklist-style rubrics instead of traditional preference rankings to train language models on subjective tasks. The framework achieved up to 28% improvement on medical reasoning benchmarks by decomposing response quality into interpretable criteria like factual accuracy and completeness.

NVIDIA AND ALPHABET VC ARMS BACK VIBE CODING STARTUP LOVABLE AT $6.6 BILLION VALUATION

Alphabet and Nvidia have invested in Swedish vibe coding startup Lovable in a $330 million Series B round that places it at a $6.6 billion valuation. Lovable has raised over $500 million this year. The company has built a product popular with both enterprises and founders. Its platform uses AI models to help users build apps and websites using text prompts.

THE SIGNATURE FLICKER

Anthropic has fixed Claude Code's signature flicker. Terminals weren't really designed for interactivity. Repositioning the cursor and writing over existing text easily leads to flickering if not done well. Anthropic chose to re-render only the changed parts. It rewrote the renderer from scratch while still keeping React as the component model.

CONTRA DSPY AND GEPA

Trying to treat LLM workflows as modular programs is a backwards, rigid, and the wrong fit for the most interesting tasks.

SUPERPOWERS 4

Superpowers 4.5 has better subagent-driven development.

PROJECT VEND: PHASE TWO

Anthropic's AI shopkeeper experiment finally became profitable after upgrading to Sonnet 4.5, adding a CRM, and hiring an AI CEO.

2026 VIBE CODING TOOL COMPARISON

Replit is the most feature-rich, well-thought-out, and powerful vibe coding tool, but v0 is the best if you're already a developer and want a technical interface.

META'S YANN LECUN TARGETS $3.5 BILLION VALUATION FOR NEW AI STARTUP, FT REPORTS

Alexandre LeBrun, the founder of French health tech startup Nabla, will be the chief executive of the new LeCun's startup.

Try the world's best coding agent, free →

GEMINI 3 FLASH

Google has introduced Gemini 3 Flash, a fast, lightweight model designed to deliver frontier-level intelligence with lower latency. It's built to help users quickly learn, plan, and build across a range of everyday tasks and applications.

GOOGLE WORKS TO ERODE NVIDIA'S SOFTWARE ADVANTAGE WITH META'S HELP

Google and Meta are working together on a new initiative to make Google's Tensor Processing Units better at running PyTorch. The move is aimed at weakening Nvidia's longstanding dominance in the AI computing market. TorchTPU will remove a key barrier that has slowed the adoption of TPU chips. It will make existing infrastructure built using PyTorch software fully compatible with TPUs. Google is considering open-sourcing parts of the software to speed uptake.

GROK VOICE AGENT API

xAI has opened up the voice stack powering Grok in Tesla vehicles to all developers. The API ranks #1 on Big Bench Audio with sub-1-second time-to-first-audio at $0.05 per minute (half the cost of OpenAI's Realtime API).

THE SQUISHY NUMBER BEHIND THE RISE AND FALL OF ORACLE'S STOCK

Oracle is down 43% from its September high despite reporting $523 billion in "remaining performance obligations". About $300 billion of that comes from a five-year OpenAI contract, but analysts are skeptical that OpenAI, which only has ~$20B in annualized revenue, will be able to pay. Nvidia's stalled $100 billion OpenAI investment (announced in September but still unsigned) compounds the anxiety over the circular deals plaguing AI infrastructure.

YES, AGI CAN HAPPEN – A COMPUTATIONAL PERSPECTIVE

Current models vastly underutilize hardware: DeepSeek-V3 and Llama-4 achieve only ~20% FLOP utilization during training, and inference runs at single-digit utilization because autoregressive models are bottlenecked on loading weights from memory, not compute. Models we see today are also lagging indicators, trained on last-gen hardware that's unoptimized for the massive size of modern clusters and the latest training methods.

WHEN AGENTS ATTACK: HOW AI COLLAPSES AND REBUILDS MARKETPLACE MOATS

AI agents, like those launched by OpenAI, threaten marketplace moats by taking over discovery, transactions, and supply workflows, thus collapsing traditional marketing funnels. Marketplaces must adapt by optimizing brand experiences, personalizing interactions, and integrating supply to maintain value. Founders should focus on high-frequency markets, fortify customer connections, and explore new monetization strategies like subscriptions or data licensing to stay competitive.

Learn more with this AI Use Case Discovery Guide.

WHAT ACTUALLY IS CLAUDE CODE'S PLAN MODE?

Plan mode is a feature in Claude Code that pauses execution until users approve of a plan. A plan in Claude Code is a markdown file written into Claude's plans folder by Claude in plan mode. Users can review and edit the plan before Claude executes whatever task it needs to. The tool brings in more complexity into the user interface, taking away some of the magic. It replicates something that natural language can already do, making the user experience a little less natural.

BUILD INTERACTIVE MINI APPS WITH OPAL IN THE GEMINI APP

Opal is a tool for building AI-powered mini apps. It is now directly available in the Gemini web app. Users can use it to create experimental Gems to unlock even more customized Gemini experiences. The tool can be found in the Gems manager.

CHATTERBOX

Chatterbox is an open-source state-of-the-art text-to-speech (TTS) model with multilingual support, emotion control, and zero-shot voice cloning.

JASSY TAPS 27-YEAR AMAZON VETERAN TO RUN AGI ORG, WHICH IS NOW DEFINITELY A THING THAT EXISTS

Peter DeSantis, SVP of Utility Computing at AWS, is taking two of his teams from AWS and putting them into a unit called 'AGI'. The unit will report directly to Andy Jassy, Amazon's Lead Product Market for GenAI and CEO. AI and AI-related compute is now an all-Amazon thing, not just an AWS thing. The unit is building things that are so important that Amazon is taking one of its most senior AWS leaders and putting him in charge of all of it.

INSIDE CHINA'S 'MANHATTAN PROJECT' TO RIVAL THE WEST IN AI CHIPS

Chinese scientists have built a working prototype of an EUV lithography machine, the $250 million tool that only ASML knows how to make and which the US has blocked from being sold to China since 2018. A team of former ASML engineers reverse-engineered the machines in a high-security Shenzhen lab, with recruits given fake IDs and signing bonuses up to $700,000. The prototype hasn't yet produced working chips, but sources expect it will by 2028-2030.

mechanism. See the exact math from ngrok

WITH FREEBIES, OPENAI, GOOGLE VIE FOR INDIAN USERS AND TRAINING DATA

ChatGPT's daily active users in India surged 600% year-on-year to 73 million, more than double its US base, after OpenAI made its Go plan free for a year.

OPERA LAUNCHES NEON BROWSER GLOBALLY WITH PAID EARLY ACCESS

Opera has globally launched its Neon browser with paid early access at $19.90/month, targeting AI power users.

OPENAI IN TALKS WITH AMAZON ABOUT INVESTMENT THAT COULD EXCEED $10 BILLION

In another circular deal, OpenAI is looking for a $10 billion investment from Amazon alongside an agreement to use AWS' Trainium chips.

Protect your app today →

GPT-IMAGE-1.5

OpenAI has launched GPT-Image-1.5, a faster and more accurate image generation model with improved instruction-following and editing. The rollout follows competitive pressure from Google's Gemini 3 and Nano Banana Pro models, which recently led key benchmarks.

SAM AUDIO

Meta has introduced SAM Audio, a new model capable of isolating specific sounds from complex audio using text, visual, or time-based prompts. As part of the Segment Anything family, it brings flexible, prompt-driven sound editing to tasks like removing background noise or isolating instruments in a recording.

OPENAI ROLLS BACK CHATGPT'S MODEL ROUTER SYSTEM FOR MOST USERS

OpenAI quietly abandoned the automatic model router, a key feature of its GPT-5, for Free and Go tier users, defaulting them to GPT-5.2 Instant instead of routing complex queries to reasoning models. The router had increased free users' reasoning model usage from under 1% to 7% but a source says it hurt daily active users because people don't want to wait 20 seconds for a better answer.

INFERENCE ECONOMICS 101: RESERVED COMPUTE VERSUS INFERENCE APIS

The infrastructure layer is no longer converging on a single 'best' model. Instead, it is splitting into two durable and economically attractive markets. One side provides reserved and hourly compute platforms designed to deliver predictability, control, and determinism for customers willing to operate infrastructure directly. The other side operates inference APIs, trading some control for scale while absorbing utilization risk and abstracting complexity in exchange for cost-efficiency and speed. Understanding the split is essential to evaluating where value will accrue as inference becomes ubiquitous.

PROMPT CACHING: 10X CHEAPER LLM TOKENS, BUT HOW?

Catched input tokens are currently 10 times cheaper than regular input tokens for both OpenAI and Anthropic's APIs. They can even reduce latency by up to 85% for long prompts. It is clear that these providers aren't just saving responses and reusing them. This post takes a look at what's actually cached and how prompt caching works in general.

IS RESUMABLE LLM STREAMING HARD? NO, IT'S JUST ANNOYING

The state-of-the-art in large language model streaming is surprisingly bad. Both Gemini and Claude require users to refresh the page if their conversation is interrupted mid-stream. This guide shows readers how to implement good, resumable LLM streams. This allows users to refresh tabs mid-stream, switch between chats mid-stream, navigate to other sections of an app without interrupting a stream, and continue streams through momentary internet connection disruptions.

the usual operational headaches. Free live stream: Tune in →

OPENAI LAUNCHES APPS SDK WITH MONETIZATION FOR CHATGPT

Developers can now build apps that live inside ChatGPT and monetize them through external checkout or, for select partners, Instant Checkout that handles payments without leaving the chat interface.

QWEN-DOC FOR LONG-CONTEXT DOCUMENTS

Qwen-Doc is an open-source initiative for Document AI, offering models and tools to improve long-context understanding, parsing, and agent-based reasoning for complex documents.