Read Online | Sign Up | Advertise
Good morning, {{ first_name | AI enthusiasts }}. Fresh off yesterday’s news of OpenAI claiming gold-level performance at the International Math Olympiad, Google entered the spotlight—earning an official gold-medal standard… and stirring up some serious drama.
The two AI giants took different approaches to generating and grading math proofs, but one thing is clear: the race to build mathematical superintelligence is officially on.
In today’s AI rundown:
Google’s ‘official’ gold win at IMO
Alibaba’s Qwen3 takes open-source crown
Create an AI agent that drafts emails for you
Brain-inspired Hierarchical Reasoning Model
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
🥇 Google’s ‘official’ gold win at IMO
Image source: Google DeepMind
The Rundown: Google DeepMind has announced that its advanced version of Gemini with Deep Think has officially achieved gold-medal level performance at the International Mathematical Olympiad 2025, following OpenAI’s similar claim.
The details:
DeepMind said it worked with IMO to test Gemini's mathematical reasoning on the same problem statements and time limits, 4.5 hours, as human competitors.
Out of six problems covering algebra, combinatorics, geometry, and number theory, the AI solved five and scored 35/42—marking the gold-medal standard.
Last year, DeepMind won silver by using domain-specific translations, but this year, its model tackled the problems entirely in natural language end-to-end.
OpenAI also claimed the same score with an unnamed model, but it did not work with IMO and had the answers graded by former medalists.
Google’s answers, on the other hand, were officially graded and certified by IMO coordinators using the same internal criteria as for student solutions.
Why it matters: Despite taking different paths, both models’ performance shows that AI is rapidly closing in on advanced mathematical reasoning. At this rate, the next frontier isn’t if they’ll solve all 6 out of 6 IMO problems—but rather when they’ll have the creativity to solve problems no human ever has.
TOGETHER WITH VANTA
🛡️ Your shortcut to startup compliance
The Rundown: Early compliance doesn’t just protect data — it also unlocks new customers and speeds up growth. Vanta’s Compliance for Startups Bundle makes it easy, with automation across 35+ frameworks and practical resources to guide you at every step.
The Compliance for Startups Bundle includes:
Step-by-step compliance checklists
Case studies from fast-growing startups
On-demand videos with industry leaders
ALIBABA
⚙️ Alibaba’s Qwen3 takes open-source crown
Image source: Qwen
The Rundown: Alibaba’s Qwen team just took the open-source crown with the release of an updated, non-thinking Qwen3 model that beats Kimi K2 across the board and challenges top closed-source models like Anthropic’s Claude Opus 4.
The details:
Following community feedback, Alibaba separated its hybrid thinking approach, training instruct and reasoning models independently.
The new non-thinking version activates 22B of 235B parameters with a 256K-context window, delivering significant performance gains.
In benchmarks, it surpassed Moonshot AI’s recently released Kimi K2 and challenged closed frontier models like Claude Opus 4 and GPT-4o-0327.
The updated model is 100% open-source and is also available as the free default model on Qwen Chat, Alibaba’s ChatGPT competitor.
Why it matters: Another Chinese team has outshined frontier labs through bold open-source innovation, despite chip constraints from the West. The achievement spotlights China’s growing dominance in AI innovation—driven not just by technical prowess, but by a strategic push for openness and global influence.
AI TRAINING
🤖 Create an AI agent that drafts emails for you
The Rundown: In this tutorial, you’ll learn how to build an intelligent AI agent that draft’s emails for you using xAI’s Grok 4 model through n8n's workflow automation platform.
Step-by-step:
Add a n8n chat message trigger and connect an AI Agent node to create your workflow foundation
Configure xAI Grok Chat Model (Grok-4-0709) with your API credentials
Add a Simple Memory node and set a Gmail integration to Create Draft (see image above as reference)
Test with: “Draft an email to john@company.com asking for meeting availability” and customize with system messages
Pro tip: Use draft mode first to review AI-generated emails are on par with your writing before switching to automatic sending.
PRESENTED BY ASAPP
🤝 Elevate customer trust with GenerativeAgent
The Rundown: ASAPP’s GenerativeAgent is an enterprise-grade AI agent that resolves real customer issues (on its own) across voice and chat. With new features focused on precision and trust, it delivers reliable, high-quality service at every step, while enabling fast deployment and instant value.
With GenerativeAgent, you can:
Drive better, smarter outcomes with AI that learns from human expertise over time
Maintain full visibility with tools to flag anomalies, track trends, and enforce compliance
Confidently launch with AI behavior testing in simulated environments
SAPIENT INTELLIGENCE
🧠 Brain-inspired Hierarchical Reasoning Model
Image source: Sapient Intelligence
The Rundown: Sapient Intelligence introduced Hierarchical Reasoning Model, a brain-inspired open-source AI that delivers unprecedented reasoning power on complex tasks like ARC-AGI and Sudoku, with just 27M parameters.
The details:
HRM’s architecture uses three principles seen in cortical computation: hierarchical processing, temporal separation, and recurrent connectivity.
A high-level module handles abstract planning while a low-level one executes fast, detailed tasks, switching between automatic and deliberate reasoning.
The approach enabled the model to beat larger ones like Claude 3.7, DeepSeek R1, and o3-mini-high on ARC-AGI 2 and complex Sudoku and maze puzzles.
With no pretraining or CoT, it points to a new kind of efficient intelligence that doesn’t need immense training data or suffer from brittle task decomposition.
Why it matters: As AI moves to real-world decision-making—efficient, brain-inspired models like HRM signal a shift toward intelligence that’s not just powerful, but also deployable in low-data environments. Sapient is already putting this into practice, helping teams with rare-disease diagnostics and pushing climate forecasting accuracy.
QUICK HITS
🛠️ Trending AI Tools
📝 Gemini Code Assist - Google’s AI coding assistant, now with agent mode
🤖 SOLO - Trae’s all-in-one Context Engineer for full software development
🧠 Composite - Turn your existing browser into an AI agent
⚙️ GEN - Create AI characters that build social media audiences end-to-end
💼 AI Job Opportunities
⚙️ Sanctuary AI - Mechanical Engineer, Automation
✉️ UiPath - Email Marketing Manager
📊 Meta - Data Scientist
🏙️ Metropolis - Market Leader
📰 Everything else in AI today
Cohere Labs introduced Catalyst Grants Program, providing free access to its models to teams tackling challenges in areas like education, healthcare, and climate.
AI video company Pika announced a new AI-only social video app, built on a highly expressive human video model, with early access waitlist now open for iOS users.
OpenAI’s ChatGPT now gets over 2.5B daily requests (meaning 912.5B annually), with 330 million coming from users based in the U.S alone.
Netflix said it used generative AI in an Argentine TV series and completed its VFX sequence “10 times faster” than it could have been completed with traditional tools.
Elon Musk’s xAI poached Ethan He, one of Nvidia’s top AI researchers who led the work on Cosmos, the company’s SOTA world model.
Runway announced its Act-Two motion capture model is now available via the API, allowing users to integrate it directly into their apps, platforms, and websites.
COMMUNITY
🎥 Join our next live workshop
Check out our last live workshop with Dr. Alvaro Cintas, The Rundown’s AI professor, and learn how to use Perplexity Comet (and other alternatives) to automate your browsing experience.
Watch it here. Not a member? Join The Rundown University on a 14-day free trial.
See you soon,
Rowan, Joey, Zach, Alvaro, and Shubham—The Rundown’s editorial team
