Welcome, humans.

So apparently Kling 3.0 just launched with a new Multi-Shot feature that's turning heads (3.9M views and counting). AI filmmaker PJ Ace used it to recreate the opening sequence of Brandon Sanderson's The Way of Kings in two days, and it looks shockingly close to a real film.

The feature lets you upload a single image, set shot lengths, and describe camera angles for each cut, giving you near-perfect continuity across an entire sequence. Speaking looks natural, clips run up to 15 seconds, and the whole thing is making Hollywood executives nervously refresh their LinkedIn profiles.

PJ's math: if a 90-second sequence took two days, a 90-minute feature film takes 180 days. Someone get this man a really good GPU and let him cook.

For our part, the last time we heard the phrase “multi-shot,” we were in Cabo San Lucas, and the results were decidedly NOT photorealistic. Decidedly not standing upright, either.

Here’s what happened in AI today:

• Anthropic released Claude Opus 4.6 with agent teams.

• OpenAI released GPT-5.3-Codex with faster performance and computer use.

• OpenAI also launched Frontier to deploy AI agents across tech stacks.

• Amazon projected $200B in 2026 AI capex, the most of any tech company.

Don’t forget: Check out our podcast, The Neuron: AI Explained on Spotify, Apple Podcasts, and YouTube — new episodes air every week on Tuesdays after 2pm PST!

Anthropic and OpenAI Both Dropped Their Best AI Models Today. Here's What Matters.

FULL COMPARISON: Read the full version on the website here.

You know those scenes in action movies where two fighters draw their weapons at the exact same time? That happened today in AI, except the weapons are language models and the stakes are… your entire workflow.

Yesterday, Anthropic released Claude Opus 4.6, and OpenAI released GPT-5.3-Codex. Same day. Zero coincidences. Definitely giving SuperBowl showdown vibes. Can we call this the “AI Bowl”? Either way, both companies are racing toward the same end zone: AI that can do your actual job, not just answer your questions.

Here's what each one brings to the table:

Firs, Claude Opus 4.6 is Anthropic's smartest model yet:

• The headline feature? A 1 million token context window (that's roughly 1,500 pages of text).

• It can now read and reason across entire codebases, massive document sets, or basically a small library, without losing track.

• It also introduces "agent teams" in Claude Code, where multiple AI agents split up tasks and work in parallel. Think of it like hiring a whole department instead of one intern. 

• Oh, and Claude now works inside PowerPoint, so you can build decks without leaving the app.

  • Corey had it make a sick one about OpenClaw in yesterday’s live!

On benchmarks, Opus 4.6 outperforms GPT-5.2 by about 144 Elo points on real-world work tasks (finance, legal, etc.), and scores highest on Humanity's Last Exam, a tough multidisciplinary reasoning test.

Oh, and in case this is your first time using Claude (or you just need a lil’ bump 😉 ), Anthropic offered Pro and Max subscribers a free $50 extra usage credit to celebrate the launch of Claude Opus 4.6, claimable through February 16—turn it on here.

As for GPT-5.3-Codex: 

• It's leaning hard into coding and computer use.

• The model is 25% faster than its predecessor, scores state-of-the-art on SWE-Bench Pro (real-world software engineering), and can now use a computer like a human, scoring 64.7% on OSWorld (up from 38.2%).

• The wildest part? GPT-5.3-Codex helped build itself, with early versions debugging their own training runs.

OpenAI also committed $10M in API credits toward AI-powered cybersecurity defense.

The big picture: It seems like, in this release anyway, that Anthropic is betting on breadth (office tools, massive context) and OpenAI is betting on depth (autonomous coding and computer operation).

Who’s the real winner here? In our opinion: anyone who learns to use both. As they say… por que no los dos? 

FROM OUR PARTNERS

Giving OpenClaw Your Credit Card? Give It a Privacy Card Instead.

AI agents like OpenClaw can book flights, buy domains, and rack up API costs—all while you sleep. But one misinterpreted prompt can drain your account or rack up charges you never approved.

Privacy virtual cards let you give your AI agent real spending power without the risk. Let it make purchases on your behalf with spending limits built into the card. Set hard limits, lock cards to specific merchants, and pause instantly if something goes sideways.

Whether you're running agentic workflows or just want smarter spending control, Privacy gives you the confidence to let your AI operate autonomously.

Sign up at privacy.com

Prompt Tip of the Day

Stop using the wrong AI tool. Productivity expert Jeff Su breaks down when to use which AI tool based on each one's specific superpower:

• ChatGPT: Most obedient—follows complex multi-step instructions without dropping the ball (perfect for detailed checklists).

• Gemini: Multimodality king—the only tool that natively handles video + audio + images + text simultaneously with a massive 1M token context window. (Also check out Jeff's Gemini tutorial)

• Claude: Best first drafts—especially for code (works on first try) and polished writing.

• Perplexity: Speed search—grabs specific facts fast while ChatGPT handles the thinking.

• NotebookLM: Zero hallucinations—only answers from your uploaded sources.

He also mentions these other, more niche tools that he doesn’t use every day:

• Grok: Its superpower is direct access to the Twitter/X firehose for analyzing breaking news in real-time, which he doesn't need.

• Gamma: for presentations.

• Eleven Labs: for voice cloning.

• Zapier or n8n: for automation (this could also be done with Opal if using Google integrations, or ChatGPT Agent kit, which everyone kinda forgot?).

• Napkin AI: for quick visuals.

Jeff's actual workflow: ChatGPT/Gemini for ideation and research → Claude for final polish → Perplexity to verify facts → NotebookLM to check accuracy.

Our favorite insight: Don't use all five tools just because they exist. Master one tool first (Jeff recommends ChatGPT, but you can start with Claude or Gemini), then add specialists only when you hit a specific problem they solve better.

FROM OUR PARTNERS

Editor’s Pick: Superinbox

AI is everywhere, but you still spend hours in your inbox. This AI learns about your conversation history and prepares ready-to-send drafts, directly in your Gmail or Outlook. Perfectly aligned with your tone of voice, habits, and preferences. No app switching, no prompt, no copy-pasting, just AI that works where you already work. Stop drowning in your inbox → start vibing through it today.

Treats to Try

1. OpenAI Frontier (also launched today) connects your CRMs, databases, and tools so you can deploy AI agents that execute tasks across your entire tech stack, managing them like employees with permissions and performance reviews (more on this another time).

2. Voxtral Transcribe 2 transcribes 13 languages with speaker diarization and word-level timestamps, streaming at sub-200ms latency for real-time voice agents ($0.003/min batch, $0.006/min realtime—realtime model open-sourced Apache 2.0).

3. Gizmo is a TikTok for interactive, vibe-coded mini apps where you scroll through genAI tools instead of videos (read more).

4. Fibr AI automatically rewrites your website's headlines and swaps images for each visitor based on which ad they clicked.

5. OpenArt gives you access to the new Kling 3.0 to create 15-second multi-shot videos where your characters stay identical across every camera angle, with auto-generated dialogue in multiple languages.

6. This blog from Tim Plaisted explains how to connect a local model to your Claude Code your quota runs out.

Around the Horn

1. Amazon projected $200B in 2026 capital expenditures — leading all Big Tech in the AI spending race — with Google close behind at up to $185B, Meta at up to $135B, and Microsoft at roughly $150B, though investors punished all of them for the massive outlays.

2. Meta began testing a standalone app for Vibes, its AI-generated short-form video feed, positioning it as a more direct competitor to OpenAI's Sora app.

3. Reddit said its AI-powered search engine could be its next major revenue driver, revealing that weekly active users for its AI-powered Reddit Answers feature grew from 1 million to 15 million over 2025.

4. Elixir creator José Valim explained why the language topped a Tencent study of 20 programming languages for AI code completion — citing immutability, first-class documentation, and a decade of ecosystem stability as key advantages for LLM reasoning.

FROM OUR PARTNERS

Wispr Flow turns your speech into clean, final-draft writing across email, Slack, and docs. It matches your tone, handles punctuation and lists, and adapts to how you work on Mac, Windows, and iPhone. Start for free today.

Intelligent Insights

• I miss thinking hard — A meditation on what we lose when AI handles our cognitive load.

• AI is Killing B2B SaaS — A founder's breakdown of why vertical SaaS companies are most vulnerable to AI disruption.

• John Hwang on how AI is eating consulting — gotta subscribe to read the full thing, but even the non-paywalled part is interesting.

• AI is just the new monoculture — this piece is about how AI has adopted our collective trend towards familiarity over novelty, but it applies to writing and content and, well, pretty much everything else too.

• Sam Altman on TBPN — he explains how GPT-5.3 Codex helped build itself, why your next job skill is managing teams of AI agents, and how Sora's killer use case turned out to be sending memes in group chats.

• Elon's three-hour Cheeky Pint — Dwarkesh Patel and John Collison cover orbital data centers, Optimus robots, China's manufacturing lead, and the most detailed breakdown of AI's hardware bottleneck we've heard from anyone.

• Steve Yegge's “Welcome to Gas Town” — The legendary engineer on what happens when AI democratizes coding. A wild idea that’s sort of the opposite of OpenAI’s Frontier; could the perfect AI interface be more like an RTS game? Anthropic’s Sholto Douglas doesn’t think so; and he even built one!

A Cat’s Commentary

P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.