In partnership with
Hey folks,
Anthropic just made a bunch of paid Claude features free and the timing is no accident. Free users can now create files (Excel, PowerPoint, Word, PDFs), connect to third-party apps like Slack, Notion, Figma, and Zapier, and use Skills for domain-specific tasks. They also added conversation compaction so free chats don't hit the wall as fast. All of this dropped the same week OpenAI started showing ads in ChatGPT. Anthropic's basically saying: "We'll give you more, and we won't make you the product."
Today in AI:
→ OpenAI ditches NVIDIA for 1,000-token-per-second coding
→ China's $700M AI giveaway war just escalated
→ Prompt of the Day: Real-time prototyping with Codex Spark
Let's dive in..
⚡ OpenAI Just Ditched NVIDIA for Its Fastest Coding Model
OpenAI just shipped a production model that doesn't run on NVIDIA hardware.
GPT-5.3-Codex-Spark is OpenAI's new real-time coding model, and it runs on Cerebras's Wafer-Scale Engine, a single chip the size of a dinner plate with the largest on-chip memory of any AI processor. The result is over 1,000 tokens per second. That's 15x faster than the full GPT-5.3-Codex.
Speed changes how you work. Instead of sending a task to an agent and waiting 15 minutes, you're having a conversation with your code in real time. You can interrupt it mid-generation, redirect it, and iterate instantly. OpenAI calls it a "daily driver" and less about solving complex autonomous tasks, more about making every small edit feel instant.
The benchmarks back it up:
SWE-Bench Pro: 56.8% accuracy, completing tasks in 2-3 minutes vs. 15-17 minutes for full Codex
Terminal-Bench 2.0: 77.3%, matching the full GPT-5.3-Codex score at a fraction of the time
Infrastructure: 80% reduction in per-roundtrip overhead, 50% cut in time-to-first-token
It's a smaller model—performance lands between GPT-5.3-Codex and GPT-5.1-Codex-Mini. You wouldn't use it for a massive codebase rewrite. But for prototyping, quick edits, refining UI layouts, and answering questions about your code? It's the fastest thing I've used.
The NVIDIA angle is the real story here. OpenAI's Head of Industrial Compute, Sachin Katti, called Cerebras "a great engineering partner" and said they're bringing "wafer-scale compute into production." This is the first crack in NVIDIA's monopoly on frontier AI inference. If Cerebras can deliver this speed at competitive cost, every cloud provider will want in.
For your team: If your developers spend more time waiting for AI responses than coding, Codex Spark is worth trying. Available now as a research preview for ChatGPT Pro users in the Codex app, CLI, and VS Code extension.
Together with Neo
Trust-First AI, Built Into Your Browser
Agentic workflows are everywhere. Real trust is still rare.
Norton Neo is the world’s first AI-native browser designed from the ground up for safety, speed, and clarity. It brings AI directly into how you browse, search, and work without forcing you to prompt, manage, or babysit it.
Key Features:
Privacy and security are built into its DNA.
Tabs organize themselves intelligently.
A personal memory adapts to how you work over time.
This is zero-prompt productivity. AI that anticipates what you need next, so you can stay focused on doing real work instead of managing tools.
If agentic AI is the trend, Neo is the browser that makes it trustworthy.
Try Norton Neo and experience the future of browsing.
🇨🇳 China Just Dropped Four Frontier Models in One Week
China's biggest tech companies turned the Lunar New Year holiday into an all-out AI arms race while Alibaba committed $432 million just to promote its Qwen chatbot. Alibaba's Qwen giveaway at bubble tea shops got so popular it caused massive queues and topped the App Store charts.
But here's the part that matters more than the marketing stunt: the models themselves are getting seriously good.
GLM-5 is a 744-billion parameter model trained entirely on Huawei chips. It’s beating Gemini 3 Pro and GPT-5.2. It first appeared as a mystery model called "Pony Alpha" that blew up on OpenRouter before anyone knew who made it
Kling 3.0 by Kuaishou generates photorealistic video up to 15 seconds with native audio across multiple languages
Qwen-Image 2.0 from Alibaba handles the image generation side, and the Qwen model is now powering AI tools for the 2026 Milan Winter Olympics
Seedance 2.0 from ByteDance generates multi-shot film sequences in roughly 60 seconds
Google DeepMind's boss told CNBC that Chinese AI models are just "months" behind Western rivals. The gap is shrinking fast. And Beijing actually had to step in and tell these companies to cool it, warning against "involutionary" competition. When your government has to tell you to spend less on AI, that's a signal.
For your team: If you haven't evaluated Chinese open-source models like GLM-5 or Qwen for your stack, the performance-to-cost math just changed. Especially if you're running inference at scale.
🎯 Prompt Of The Day
Use Codex Spark's real-time speed to rapidly prototype UI components. Paste this into a Codex session with your project open:
Look at the dashboard in my codebase. I want to iterate on the design quickly. First, show me the current layout structure in plain English. Then give me three alternative approaches:
(1) a minimal version that strips it to essentials
(2) a version optimized for mobile-first
(3) a version that adds [specific feature—e.g., dark mode, animation, accessibility improvements].
For each, generate the code changes I can apply immediately. Let's start with option 1 and I'll redirect you from there.
The key with Spark is treating it like a conversation, not a task queue. Give it one thing, react, redirect. That's where the 1,000 tok/sec speed actually pays off.
🐝 AI Buzz Bits
🔧 Ex-GitHub CEO launches Entire to track AI-written code. Nat Friedman raised $60M at a $300M valuation for a startup that records the prompts, reasoning, and decisions behind AI-generated code. The first product, Checkpoints, works with Claude Code and Gemini CLI so teams can actually audit what the AI did and why.
❄️ Snowflake built an AI coding agent that actually knows your data. Cortex Code actually understands your Snowflake data, governance rules, and compute setup. Talk to it in plain English and it builds production-ready data pipelines.
💰 Anthropic just raised $30 billion at a $380B valuation. That's the largest private funding round ever. Revenue hit a $14B run rate, growing 10x annually for three straight years. Claude Code alone is at $2.5B+ run rate.
🛠 Tool Spotlight
Oz — Cloud infrastructure from Warp that runs hundreds of AI coding agents in parallel. Each agent gets its own Docker environment to build, test, and write PRs.
Kilo Code — Open-source coding agent that's #1 on OpenRouter with 1.5M+ developers. Works in VS Code, JetBrains, Cursor, Windsurf, and standalone CLI. Bring any model.
Cosmic CLI — Go from idea to deployed app in one terminal command. Describe what you want, and it generates a content model, builds a Next.js app, pushes to GitHub, and deploys to Vercel.
For a full list of 1500+ AI tools, visit our Supertool Directory
👉 Know someone drowning in AI news? Forward this to them or send your unique referral link
Cheers, Tim
