Dec 15, 2025 5 min read Daily Rundown

🛎️ Open Models Break Math

Plus: Agents as APIs, No Ceiling on Pay

Good Morning, AI Enthusiasts!

Capability is decentralizing, leverage is multiplying, and the economic gravity of AI is moving fast.

LLM

Small Open Model, Big Math Signal

👀 What’s happening: Nous Research has open sourced Nomos 1, a 30B parameter reasoning model that scored 87 out of 120 on the 2025 Putnam Contest. That result would have ranked second last year among nearly 4,000 human competitors. The model uses a two-phase worker and critic tournament system and ships with an open reasoning harness. Running Qwen 3 through the same setup scored only 24, isolating the gains to model training, not orchestration.

🔥 How this hits reality: This is a shot across the bow for closed frontier labs and oversized math benchmarks. A 30B open model now posts near elite human performance on a notoriously brutal exam. That changes procurement math for research teams, universities, and applied science groups who no longer need proprietary systems to do serious reasoning work. Open weights plus open orchestration compress iteration cycles and quietly weaken the moat around premium reasoning APIs.

🛎️ Our take: When a small open model can nearly win the Putnam, reasoning stops being a scale problem and becomes a training discipline problem.

TOGETHER WITH KARAVIDEO

All AI Video Models in One Platform

There are too many great AI video models — Kling, Runway, Veo 3, Pika — but too many tabs. KaraVideo.ai fixes that. It brings all the world’s top AI video models into one place.

🎬 Generate, compare, and remix without switching tools.
🎭 Keep your characters consistent, your style intact, and your creativity flowing.
💡 One login, one dashboard, endless creation.

👉 Explore KaraVideo

NEW LAUNCH

Google Ships Research Agents as APIs

👀 What’s happening: Google released a reengineered Gemini Deep Research agent built on Gemini 3 Pro, repositioning it from a standalone research tool into a programmable system. The key change is the new Interactions API, which lets developers embed Google’s long-context, multi-step research capabilities directly into their own applications. Google is also rolling the agent into Search, Finance, NotebookLM, and the Gemini app, signaling a march toward agent-driven information workflows.

🔥 How this hits reality: This turns deep research into infrastructure. Instead of analysts or knowledge workers manually querying tools, teams can now hardwire research agents into diligence platforms, regulatory reviews, scientific analysis, or internal knowledge systems. It pressures research AI vendors whose value stops at their own chatbot, and it favors platforms that can guarantee factual stability over long-running tasks. Google quietly strengthens its cloud and ecosystem lock-in through reliability, not features.

🛎️ Our take: When research becomes callable, not clickable, the advantage shifts to whoever controls agent reliability at scale. Google just made its bet explicit.

TALENT

AI Talent Competition Keeps Escalating

👀 What’s happening: OpenAI has removed its six month equity vesting cliff, allowing stock to vest immediately for new hires. The move follows earlier cuts from the one year norm and mirrors changes at rivals like xAI. As Meta, Google, and Anthropic chase the same narrow pool of elite researchers, traditional retention mechanisms are being stripped away. This is less about generosity and more about keeping pace in a tightening labor market.

🔥 How this hits reality: Talent competition is intensifying, and compensation is absorbing the pressure. Companies are trading structure for speed, accepting dilution and churn to avoid hiring stalls. Managers plan for shorter tenures, finance teams normalize outsized equity spend, and recruiting becomes a continuous arms race. Smaller labs lose ground quietly as they cannot escalate fast enough. The market is converging on whoever can pay and move fastest.

🛎️ Our take: As the AI race accelerates, talent competition compounds itself. Every concession raises the floor, and no serious player can afford to step off.

QUICK HITS

Google Translate is adding real-time headphone translations along with Gemini-powered translation and learning upgrades.
New York passed new laws requiring disclosure of AI-generated people in ads and consent for commercial use of deceased individuals’ likenesses.
OpenAI expects to launch an adult mode in ChatGPT in Q1 2026, pending improvements to its age-prediction and safeguarding systems.
Thinking Machines has launched Tinker as a generally available LoRA-based fine-tuning service for large models.
Google is bringing basic Gemini AI features to Chrome on iPhone and iPad for on-page questions and summaries.

TRENDING

Daily AI Launches

Qu has launched a game-based platform that turns Electronics and Physics concepts into interactive puzzles for STEM learning.
BlazorOcticons has launched a library offering GitHub Octicons as native, customizable Blazor components.
Flowglad has launched an open-source payment provider that eliminates webhooks and glue code for database syncing.
Google Vids has launched an AI video tool for work that creates polished videos with storyboards and avatars in minutes.

Featured AI Tools

🎥 KaraVideo unites all AI video models in one place.
📚 Heardly is the Fast Way to read Best Book.
🪶 CopyOwl is the First AI Research Agent, deep research on any topic in one click.
🤖 Momen AI bridges the gap from vibe to viability, powering Lovable apps without code.

TOGETHER WITH US

AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.

We operate the industry’s most influential portfolio of newsletters, each shaping a different frontier of the AI & Tech revolution:

Be Smarter in 5 Minutes