Open-source AI crushes elite math exam

PLUS: Fix bugs and ship features from Slack with Claude Code

Good morning, AI enthusiasts. Remember when AI couldn't handle basic arithmetic? Now, a small open-source model is crushing one of the world’s hardest math exams.

After achieving top scores in the Putnam Contest, Nous Research’s Nomos-1 joins a long list of AI-driven math advances this year that show the field may be about to move into a whole new territory of discovery.

In today’s AI rundown:

  • Nous Research's AI takes on elite math exam

  • Microsoft maps how people use Copilot

  • Fix bugs and ship features from Slack with Claude Code

  • AI ring gives ‘external memory’ for your brain

  • 4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

NOUS RESEARCH

Image source: Nano Banana Pro / The Rundown

The Rundown: Nous Research just open-sourced Nomos 1, a new 30B parameter reasoning system that scored 87 out of 120 on the 2025 Putnam Contest — crushing rivals like Qwen 3 on one of the most prestigious collegiate math competitions.

The details:

  • The system uses a two-phase approach: AI ‘workers’ solve and self-critique responses, with a tournament-style bracket then selecting the best submission.

  • Nomos’ score would have placed second among nearly 4,000 human competitors last year, with the model earning eight perfect problem scores.

  • Nous also released and open-sourced a reasoning harness — orchestration code that manages how the model solves problems.

  • Running Qwen3 through the same harness and setup scored just 24/120, with the result showing gains coming from model training rather than the harness.

Why it matters: Not too long ago, even simple math problems were an issue for top AI systems —  and now, a small, open model is taking down a notoriously difficult exam. Between Nomos, AI helping conquer unsolved problems, and labs coming with gold medal-winning math models, the entire field looks ready for an AI-driven boom.

TOGETHER WITH VANTA

The Rundown: AI has unlocked new velocity for startups — and new visibility, too. The faster you grow, the sooner you’ll need to prove you’re secure enough to play with enterprise customers. Join Vanta for a live session on how to make compliance work at your pace, without slowing momentum, stalling deals, or putting revenue at risk.

You’ll walk away with:

  • Tips on identifying the right framework(s) for your startup

  • Advice on how to work security into your budget

  • Security best practices to implement now so you’re audit-ready next year

  • Essential steps to earn customer trust (and deals)

AI RESEARCH

Image source: Microsoft

The Rundown: Microsoft just published new research analyzing 37.5M Copilot conversations from the past year, revealing distinct behavioral patterns in how users engage with the AI assistant across different devices, time periods, and topics.

The details:

  • Health and wellness queries dominated mobile use regardless of hour or month, positioning phones as around-the-clock personal wellness companions.

  • Advice-seeking grew throughout the year, with users increasingly treating AI as a guidance source rather than just a pure search tool.

  • Late-night sessions saw philosophical, religious, and existential topics climb in popularity, while relationship chats spiked specifically around Valentine's Day.

  • Programming dominated in January, while social topics rose later in the year, reflecting a shift from early adopters toward a broader, mainstream audience.

Why it matters: There has been a wealth of data from major labs on how users are leveraging AI, but this Microsoft study gives an interesting look at the shifting dynamics that occur based on both the time of day and year, and the device being used — insights that can shape how next-gen assistants adapt and optimize for context.

AI TRAINING

The Rundown: In this tutorial, you will learn how to add Claude Code to your Slack and task the autonomous coding assistant with fixing bugs or implementing new features, without ever opening a code editor.

Step-by-step:

  1. Connect Claude Code to GitHub here and add the Claude app to your Slack workspace here.

  2. You should see it in the bottom left of your Slack now. Click “Connect Account” and give Claude access to Slack.

  3. Add Claude to an existing channel or to a new one by typing “@claude” and hitting enter. It will ask you to add Claude to the channel. Approve it.

  4. You can now @Claude and give it coding tasks, with the assistant building context based on recent messages.

Pro tip: Start a Slack thread on a particular issue for the best performance. Additionally, you can also now access Claude Code on mobile via the Slack app.

PRESENTED BY GURU

The Rundown: Guru is the AI Source of Truth that connects all of your company’s tools and delivers cited, permission-aware answers everywhere you work. With one governed knowledge layer powering both your people and your AIs, teams move faster — with fewer blind spots and mistakes.

Guru allows you to:

  • Connect all knowledge with permission-aware access

  • Get trusted, cited answers in chat and everywhere else you work

  • Experience knowledge that improves and verifies itself

AI WEARABLES

Image source: Core Devices

The Rundown: Pebble maker Core Devices just introduced the Index 01, a new $75 AI smart ring voice recorder that captures spoken ideas and uses on-device AI to turn them into notes, reminders, or calendar entries, without subscriptions or internet.

The details:

  • The ring fits on a user’s index finger with a thumb-activated button, allowing for hands-free recording while on the move.

  • Recordings sync to a user’s phone, where a local LLM transcribes and processes the voice note via an open-source speech-to-text system.

  • The ring requires no charging, with batteries lasting up to two years of typical use, and can record up to five minutes of continuous audio.

Why it matters: After wearables like the Humane Pin and Rabbit R1 stumbled trying to replace phones with AI hardware, Index 01 is taking a narrower approach — a single, simple task executed reliably. It’s certainly no guaranteed success, but it might reveal whether the device market has room for focused tools vs. mass-market moonshots.

QUICK HITS

  • 🧑‍💻 Devstral 2 - Mistral’s next-gen family of coding-focused models

  • 💡 Stitch - Google’s tool to turn ideas into UI designs, now using Gemini 3

  • 🧮 Nomos 1 - Nous Research’s powerful AI math reasoning system

  • 🧠 Purpose - AI mentor for deep, personalized guidance on demand

Just launched: Coder now provides full-stack infrastructure for governed AI development at scale. Learn more about AI Bridge, Agent Boundaries, and Coder Tasks.*

The U.S.’ annual defense bill reportedly mandates the creation of a committee to study the military impacts of AGI and countermeasures to adversaries pursuing it.

Chinese AI startup DeepSeek is reportedly developing its next AI model using thousands of illegally imported Nvidia chips, according to The Information.

McDonald’s Netherlands pulled an AI-created Christmas ad after facing backlash, saying the reaction “serves as an important learning as we explore effective use of AI.”

Amazon and Microsoft both announced major AI and cloud infrastructure investments in India, collectively pledging over $52B.

*Sponsored Listing

COMMUNITY

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Courtney P. in Sylvan Lake, MI:

"My daughter is a freshman in high school and has big dreams of playing field hockey at the collegiate level. I created a GPT in ChatGPT that acts as a college recruiting coach for her that can give her everything from coaching, training workouts, nutrition tips, mindset coaching, to which clinics are the best to be at to get in front of coaches.

We’ve saved thousands of dollars at this point from having to hire an outside recruiter coach just by me spending a few hours to create this knowledge base for her!"

How do you use AI? Tell us here.

That's it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you.

Login or Subscribe to participate in polls.

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Reply

or to participate.