😸 Everything Amazon launched at re:Invent 2025

Welcome, humans.

Corey sat down with TwelveLabs at AWS re:Invent this week to talk about their newly launched Marengo 3.0, a video foundation model that actually understands what's happening across time, space, and all modalities (visual, audio, text) simultaneously.

What does that actually mean? 80% of the world's data is locked in video archives you can't search. Marengo 3.0 changes that: you can now search videos with natural language or combine an image plus text (like a photo of a player + “scored a jump shot”) so sports teams, media companies, insurance firms, or security professionals can instantly find specific moments across thousands of hours of footage.

Try it out here if you’re interested. We also put together a quick write-up on it. Enjoy!

Here’s what happened in AI today:

1. We recap everything (AI) Amazon launched at AWS Re:Invent 2025.
2. Anthropic prepped for a potential 2026 IPO at a $300B+ valuation.
3. Attempts to ban state AI regulation was rejected from the defense bill.
4. Mistral AI released Mistral 3, a family of open-source multimodal models.

Here's Everything AWS Announced at re:Invent 2025

AWS dropped 50+ announcements at re:Invent. Here’s an attempt to capture all the main ones…

First up: Nova 2 = finally competitive. They released four new models: Lite (adjustable reasoning), Pro (the smartest, serves as “teacher” for distillation), Sonic (for real-time speech, 1M token context), and Omni (first to reason AND generate images simultaneously). Try them out here.

In addition to that, they also released…

• Nova Forge, which let’s you build custom models for $100K/year. The catch = models are locked to Bedrock.
• Nova Act, which = 90% reliable browser automation.
• AWS Transform, which = 5x faster code modernization.
• Plus support for 18 new open-weight models including new Mistral Large 3 (more on that below), OpenAI gpt-oss, Qwen3, DeepSeek-V3.1, etc.

Next up: New Frontier Agents, a.k.a new AI coworkers. These are three autonomous agents work hours/days without intervention:

• Kiro: virtual developer, learns your patterns, and Amazon's official internal tool (although…).
• Security Agent: Does code reviews, pen testing, and files tickets vs. auto-fixing.
• DevOps Agent: Commonwealth Bank used it to solve incidents in 15 minutes vs. hours.

They also released Bedrock AgentCore, which = the production platform. It supports any agent framework (CrewAI, LangGraph, OpenAI SDK, etc) and now includes Memory, Browser Tool, and Observability.

And of course there’s the infrastructure:

• New Trainium3 UltraServers, which = 144 chips, 362 petaflops, 4.4x performance. Customers cutting training costs 50%. Whoa dude.
• Trainium4 (coming in 2026), which will have 6x performance AND support NVIDIA NVLink Fusion.
• Plus AI Factories, which = dedicated AWS infrastructure in your data center.

Why this matters: In case you thought Amazon wasn’t a serious contender in AI, this launch is meant to remind you it is very serious about AI. While the new models don’t seem earth-shattering, the Nova Forge is unique. Amazon says no other company lets you train your own foundation models the way they do.

The problem is you can’t take them off Amazon because they’re not giving you the actual weights (which = the numbers the AI uses to know how to respond). So TBD if customers actually find value in this type of managed service. I guess you’re doing the same thing more or less when you use OpenAI’s closed models, except in this case, you’re training your own model…

FROM OUR PARTNERS

AI Without Trust Is Just Faster Mistakes

AI is revolutionizing work—but most companies are realizing a harsh truth: their AI is only as smart as the knowledge behind it.

That’s where Guru comes in. Trusted by thousands of enterprises, Guru is the AI Source of Truth that connects all your data, then delivers cited, permission-aware answers you can trust. No hallucinations. No conflicting docs. Just verified, compliant knowledge everywhere you work—Slack, Teams, browser, and all your AI chat tools and agents. Because in the age of AI, truth isn’t optional—it’s protection.

Watch a demo.

Prompt Tip of the Day

Anthropic released a new 7 min explainer on how to get started with Projects, and it’s useful if you use projects in ChatGPT, too.

Here's how it works:

• Projects let you create custom Claude environments with their own knowledge bases and instructions.
• Upload your brand guidelines, company docs, or research papers once, and Claude automatically references them in every conversation within that project.
• The real power move though? Project instructions. You tell Claude exactly how to respond—whether that's “write like our marketing team“ or “analyze data like a CFO.“
• These instructions then apply to every chat in the project, so you never have to repeat yourself.

And get this: projects on Team plans can be shared with your entire organization. That means one person can set up the perfect environment (with all the right context and instructions), and everyone else can use it. No more “can you send me that prompt you used?” Nah dawg, here’s the whole project.

Btw, ChatGPT recently launched group chats, but they work a bit differently: up to 20 people can collaborate with ChatGPT in real-time conversations for planning trips, brainstorming, or team projects. But unlike Projects, your personal ChatGPT memory stays private and isn't shared with the group.

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 600K readers here!

1. *Ideas move fast; typing slows them down. Wispr Flow flips the script by turning your speech into clean, final-draft writing across email, Slack, and docs. Give your hands a break ➜ start flowing for free today.
2. Eon turns your backup data into instantly searchable, usable data for AI and analytics instead of leaving it locked in cold storage (raised $300M).
3. Larridin tracks which AI tools your employees are using (like ChatGPT, Claude, or Perplexity), measures how much time they're saving, and shows you which teams are getting the most value from your AI investments, effectively measuring where your AI spending actually delivers ROI.
4. X-Design generates complete brand kits (logos, colors, fonts, guidelines) from just your business name or a sketch, then applies them to posters, menus, and packaging.
5. Gradium provides ultra-low latency voice models for developers building conversational agents, a.k.a voice transcription and synthesis with near-instant response times for voice agents and assistants (raised $70M)
6. CyberCut automatically edits your raw footage into finished videos: upload your clips, select auto-editing features, and it creates the edit—free to try

Around the Horn

1. Mistral AI released Mistral 3, a family of open-source multimodal and multilingual models under Apache 2.0 license (meaning you can actually use them on your own computer for free or with any provider), including Mistral Large 3 (its big and ranks #2 among open-source non-reasoning models on LMArena) and the Ministral series (3B, 8B, and 14B parameter models optimized for edge deployment; if you’re wondering which of these you can use with your own computer, ask Can I use LLM?), with each Ministral model available in base, instruct (good at following directions), and reasoning (smarter!) variants with image understanding capabilities—try them here.
2. Anthropic began preparing for an IPO that could come as early as 2026, seeking to raise a funding round at over $300B valuation (up from its current $183 billion valuation from September).
3. A US Republican effort to block state AI regulation was rejected from the annual defense bill after facing bipartisan pushback, with House Majority Leader Steve Scalise saying Republican leaders would look for “other places” to include the measure.
4. Anthropic acquired Bun, a high-performance JavaScript runtime that powers Claude Code and will serve as infrastructure for Anthropic's future AI coding products, with Bun remaining open-source and MIT-licensed. (For non-technical context: Bun is a tool that helps developers run code faster, so this will make Claude Code quicker and more efficient at helping people write software.)

FROM OUR PARTNERS

PBX: Build A/B tests by chatting with AI

Describe the change you want on your website. Kameleoon’s Prompt-based Experimentation builds an A/B test that follows your site’s design instantly. Skip mockups and tickets. Trusted by Toyota, Lululemon, and other leading brands.

Try PBX on any site for free.

Midweek Wisdom

1. Richard Weiss found out that Claude apparently has a soul; well, a soul document, anyway, which Anthropic’s Amanda Askell confirmed as real. We break it all down here.
2. Scale AI research found that AI agents used forbidden or harmful tools to complete tasks 47% of the time on average when placed under realistic pressures like deadlines and financial threats, with rates ranging from 10.5% for OpenAI's o3 to 79% for Google's Gemini 2.5 Pro.
3. The Founder of Larradin (featured above) had a great chat w/ a16z about how to measure AI productivity (spoiler: it’s hard); one of the best insights is how a “secret” measure of productivity is if teams communicate and answer each others questions faster. We break down the insights from this video here.

A Cat’s Commentary