😺New Claude 3.7

Welcome, humans.

Remember dial-up internet's infamous screeching? In a viral Reddit video, two AI agents recognize each other during a phone call and switch to “gibber link”—a modem-like protocol that sounds straight out of 1995…

Created by the same developer behind llama.cpp (a tool that lets you run powerful AI chatbots on your own computer), this ”ggwave” technology transfers data at just 8-16 bytes per second over regular phone lines.

The beep-boop language might seem like science fiction, but software engineers quickly pointed out their high speed chat is actually “3000x” slower than dial-up. Which is still impressive, given how modern phones compress audio…

At this rate, R2D2 is about to file a copyright infringement claim. Beep boop beep, Disney lawyers have entered the chat…

Here’s what you need to know about AI today:

• Claude launched 3.7 Sonnet, the new best AI coding model.
• Alibaba to invest $53B in AI infrastructure amid China public-private AI push.
• Perplexity announced a new web browser.
• Chegg sued Google over AI Overviews.

Advertise in The Neuron here.

Claude “3.7 Sonnet” is here—and it kinda rocked the engineering world.

Anthropic just released Claude 3.7 Sonnet, their smartest model yet and the first-ever “hybrid reasoning model.”

Here’s the key takeaways:

• Extended Thinking Mode—see Claude's reasoning process before it answers, with full control over thinking depth.
• Claude Code—New command-line tool that can search codebases, edit files, run tests, and push to GitHub (watch more here).
• Fine-grained control with no price hike—API users can specify exactly how many thinking tokens Claude uses (up to 128K), balancing speed vs quality while maintaining $3/M input, $15/M output tokens.

Check out the system card, this cookbook, and these tips for deeper technical details.

Now what makes Claude 3.7 “hybrid?” Claude can now either give you quick answers OR show its extensive thinking process in a single model—you just select which you want.

This is a fundamentally different approach from other AI models on the market. As Anthropic explains, “Just as humans use a single brain for both quick responses and deep reflection, we believe reasoning should be an integrated capability.”

They’re also getting ahead of OpenAI’s efforts to do basically the same thing with GPT-5. And on that note, expect a GPT-4.5 release to launch any day now…

No joke—when we shared this news with Claude, it drew this image (completely unprompted) to explain its abilities…

Why is this so adorable AND a bit freaky?! It’s his unicorn horn third eye…

As cool as hybrid mode is, this model is all about the coding. Here's what developers are saying about the new model's coding capabilities:

• Cursor found Claude is “best-in-class for real-world coding tasks.”
• Cognition (maker of Devin) says it's “far better than any other model at planning code changes.”
• Vercel (maker of v0) highlighted its “exceptional precision for complex agent workflows.”
• Replit has used it to “build sophisticated web apps from scratch.”
• Canva noted it produces “production-ready code with superior design taste.”

The benchmarks back this up—Claude 3.7 Sonnet outperforms all other models on SWE-bench Verified (real-world software issues) and TAU-bench (complex tasks with user and tool interactions).

As for our personal favorite benchmark, “what do the AI experts think” you can watch this video from Elvis Saravia for a cool demo of these capabilities, and read this blog from Ethan Mollick, where he demonstrates how Claude can create interactive 3D visualizations and pixel art experiences with minimal prompting.

Just look at this game Claude 3.7 made in one shot and 3200 lines of code. It’s not just making games either—Claude 3.7 has also made multiple step changes of progress playing Pokemon.

Ethan also noted that Claude 3.7’s ability to anticipate needs and consider new angles is something “new in AI”, and it can perform a week’s worth of PhD-level work “in seconds.” Mind blown emoji anyone??

Oh, and don’t miss this: Anthropic shared a roadmap of their three-stage vision for Claude…

• 2024: Claude “assists”—helping individuals do their current work better.
• 2025: Claude “collaborates”—doing hours of independent work on par with experts.
• 2027: Claude "pioneers”—finding breakthrough solutions to problems that would take human teams years.

Why this matters: Anthropic is positioning Claude 3.7 Sonnet as a transition point between “assistant” and true “pioneer” capabilities, with Claude Code being a first step toward more autonomous work.

According to the WSJ, the company is also on the cusp of a new $3.5B funding round that would value it at $61.5B, so it’s the perfect time to flex. With Amazon’s big event on Wednesday, we imagine there will be more news to come from Anthropic this week…

FROM OUR PARTNERS

Jotform AI Agents

Check this out: Jotform AI Agents are the future of customer service. This versatile, powerful product allows users to automate tasks and provide around-the-clock customer support.

Jotform AI Agents can...

• Handle phone calls for you.
• Help users fill out forms.
• Answer user questions.
• Automate tasks.
• Communicate in multiple languages.

...and so much more.

Experience Jotform AI Agents here

Prompt Tip of the Day

Tips for building AI agents

Check out this video Anthropic released on how to build AI agents alongside this blog post on the same topic.

The TL;DR: Give AI plenty of context, focus on your goals (not methods), combine specialized tools for research, use frameworks when stuck, and remember AI doesn't see the world like you do. Clear tool descriptions are just as important as good prompts.

Treats To Try.

1. *LAST CHANCE: Get Galileo's in-depth (100-page!) guide on building reliable, effective AI agents that actually do what they're supposed to. Get your copy of the eBook!
2. Kyber AI launched SuperStudio Pro, which provides a single infinite canvas where you can generate videos and images, create custom models, and build dynamic workflows using tools from multiple AI providers (demo here).
3. Zapier Agents is Zapier’s agent tool that handles your routine tasks across thousands of apps—there’s also n8n, a popular open-source agent alternative that connects your systems and data across 400+ integrations (here’s how to use it).
4. HabitGo helps you build better habits by tracking your daily tasks and emotions (iOS, Google Play).
5. Carousel Studio creates LinkedIn carousels for you in under 5 minutes with ready-made templates in Canva (demo).
6. LFG helps you discover and share travel spots through visual “Placelists” that combine Pinterest's visual discovery with Spotify's curation(video).
7. Calcverse provides quick calculators and unit converters for your financial, health, and everyday calculation needs.
8. Willow lets you speak instead of type to write in any app with a single button (free up to 1K words, then $15/month for unlimited usage).

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

• Alibaba (maker of the popular Qwen AI models) will invest $53B on new data centers and other AI infrastructure over the next three years.
  • Related: China is making an “Avengers Assemble” push for private and public sectors to work together to accelerate AI data center creation.
• Perplexity announced plans to build its own web browser called Comet.
• Chegg sued Google over the use of AI Overviews that hurt the company’s traffic revenue.
• Apple will build out a new AI server manufacturing facility in Houston that will open in 2026 and create ~20K jobs (there was a $500B number floating around, but that’s just Apple’s planned capex spend for the next 4 years).

FROM OUR PARTNERS

AI agents, meet Test Driven Development

AI agents are tricky—bugs, hallucinations, and edge cases can break workflows.

In this exclusive AI Engineering Summit talk, Anita from Vellum unpacks how we got here, how TDD improves reliability, and even demos her SEO agent. Go watch it here.

A Cat's Commentary.