Welcome, humans.

Anthropic just dropped Claude Opus 4.7, and instead of writing yet another benchmark breakdown, we're putting it through the gauntlet in real time.

Starting RIGHT NOW, join Grant Harvey (Lead Writer @ The Neuron) for an unscripted, warts-and-all live test of Anthropic's newest flagship model. No rehearsal, no cherry-picked demos, no pre-approved prompts. Just "here's what actually happens when you throw hard problems at it."

If you build on Claude, this is the model you're about to be using. It's a direct upgrade path from Opus 4.6 at the same price ($5 per million input tokens and $25 per million output tokens). It's also the first model Anthropic is shipping with their new automatic cyber safeguards, a direct follow-up to last week's Project Glasswing announcement.

🔴 What we're testing (live, on stream):

• Advanced coding with agents (on a few personal projects, and Cat DOOM of course!)

• The new higher-resolution vision (images up to ~3.75 megapixels, more than 3x what Claude used to handle)

• File system-based memory across multi-session agent work

• The new "xhigh" effort level (between "high" and "max") on hard reasoning problems

• Claude Code's new /ultrareview slash command for bug and design review

• Auto mode for longer, less-interrupted agent runs (now extended to Max users)

Bring your hardest prompts. We'll run them live and show you what breaks, what shines, and whether it's worth migrating today. Drop them in the chat during the stream or reply to this email right now; we'll work them in.

Watch live at now here:

• YouTube Live

• LinkedIn Live

• X (Twitter)

P.S. Timestamps will be added to this post after the stream, so if you can't make it live, bookmark this email and come back tomorrow; you'll be able to jump straight to the parts you care about.

🧪 What's changing under the hood (so you know what to watch for)

Here's what Anthropic shipped today that will actually change how Opus 4.7 behaves versus 4.6, and what that means for your existing prompts and agent setups:

The upgrades worth knowing:

• State-of-the-art on GDPval-AA, a third-party evaluation of economically valuable knowledge work (finance, legal, and other pro domains where AI has to actually be useful, not just impressive)

• Big jumps on Finance Agent evals, document reasoning, and long-context work, which maps pretty cleanly onto the kind of work most knowledge workers actually use Claude for

• Substantially better instruction following (important gotcha: prompts written for earlier models can now produce unexpected results because 4.7 takes instructions more literally, no more skipping steps or loosely interpreting what you meant)

• Better "memory" across long, multi-session agent work; the model is much better at using file-system-based notes

The tokenizer change (plan for this):

• Opus 4.7 uses a new tokenizer that can map the same input to 1.0-1.35x more tokens depending on content type

• You may also see more output tokens at higher effort levels, especially on later turns in agentic settings

• Translation: your Claude bill might shift slightly even at the same price per token. Anthropic published a migration guide with effort-tuning advice; we'll dig into this on stream

🔑 The bottom line: Opus 4.7 is the model you're about to be using. The question is whether the upgrade path is smooth or whether your prompts and harnesses need re-tuning. That's what we're finding out, live.

Related reads:

• Anthropic's announcement

• Opus 4.7 System Card

• Cyber Verification Program (if you're a security researcher who needs higher-capability cyber use)

• Watch Nick Saraev’s recap — very good.

📬 Full writeup in tomorrow's newsletter

Not free atm? We've got you. We'll publish a full breakdown of everything we found in tomorrow's daily newsletter, with the sharpest prompts, the biggest surprises, and a direct "should you switch?" call.

More videos you’ll li

Real quick: Want to see your AI-adjacent product or service show up right here, below these podcast promos? Click the button below to advertise to our 675K+ readers!

🎙️ In Case You Missed It…

1. Interested in what's missing before we hit AGI? Watch: This Company Mapped the Entire World in 3D. Here's Why.

TL;DW: Peter Wilczynski, CPO at Vantor (formerly Maxar), built a 3D model of the entire Earth at 50cm resolution and made it machine-readable. He argues spatial intelligence is the gap nobody's talking about in AI, and probably the missing piece before agents can actually operate in the physical world.

Why you should watch: If you've ever wondered why AI can write code and solve math olympiad problems but still can't reliably tell a drone where to go, this one answers it. Also, there's a wild bit about how the physical world becomes the new navigation layer for AI agents.

• YouTube: Watch Here

• Spotify: Listen Here

• Apple Podcasts: Listen Here

2. Want to understand why DeepSeek keeps eating everyone's lunch? Watch: AlphaGo's Co-Creator Raised $2B to Open-Source Frontier AI

TL;DW: Ioannis Antonoglou, co-founder and CTO of Reflection AI (and the guy who co-built AlphaGo at Google DeepMind), explains why every frontier open-weight model right now comes from China, how mixture-of-experts architecture makes trillion-parameter models run efficiently, and why he raised $2B to build an American answer to DeepSeek.

Why you should watch: If 80% of US startups are building on Chinese base models (and they are), you want to understand what's happening here. The mixture-of-experts explainer alone is worth the watch.

• YouTube: Watch Here

• Spotify: Listen Here

• Apple Podcasts: Listen Here

3. Want agents that actually work on real tasks? Watch: Inside the Secret Labs Where AI Learns to Work

TL;DW: Nick Heiner, VP of Product at Surge AI (a $1.2B-revenue company built without VC money), reveals why even GPT-5, Claude, and Gemini still fail about 40% of real workplace tasks, what makes a good RL environment, and his bold prediction of a $1B company with one human employee by 2030.

Why you should watch: If you're trying to get AI agents to actually finish real work (and not just demo well), this is the missing piece on why they keep falling short.

• YouTube: Watch Here

• Spotify: Listen Here

• Apple Podcasts: Listen Here

4. Interested in where AI meets actual factories? Watch: 80% of US Factories Have Zero Robots. Google Wants to Fix That.

TL;DW: Google's Intrinsic CTO Brian Gerkey co-created ROS, the open-source software running more than 1 million robots including NASA's. He explains why most US factories still have zero automation, and how generative AI is about to change that.

Why you should watch: If you think "AI and robotics" is still science fiction for your industry, this interview resets the timeline. The factory floor is where the next AI wave actually lands.

• YouTube: Watch Here

• Spotify: Listen Here

• Apple Podcasts: Listen Here

One more before you go:

Last week we went live with the "AI for Total Beginners" 5-Level Starter Stack, a full framework for going from "I've never touched AI" to "AI saves me 10 hours a week." No coding required. It's at 5,700+ views and counting.

Last thing: And if you haven’t subscribed yet, please do! Click the image below to go to our channel and hit “subscribe” to get notified right when new videos go live.

We have a goal to hit 50K subscribers by the end of the year (if not 100K), and we’re only 33K away! If you like learning about AI, and already watch some of our videos, do us a favor and click here to subscribe today.

Stay curious,

The Neuron Team

P.P.S: Love the newsletter, but don’t want to receive these podcast announcement emails? Don’t unsubscribe — adjust your preferences to opt out of them here instead.