📈 Kimi K2 Thinking Sets New Open-Source Records

Also: AI Agents Level the Playing Field for Neurodiverse Workers

Author

Evolving AI
November 10, 2025

In partnership with

Welcome, AI enthusiasts

Moonshot AI is raising the bar with its new Kimi K2 Thinking model, a trillion-parameter powerhouse built for deep reasoning and complex problem-solving. Designed as a true “thinking agent,” it’s set to redefine what open-source AI can do in coding and multi-step tasks. Let’s dive in!

In today’s insights:

  • Kimi K2 Thinking Sets New Open-Source Records

  • AI Agents Level the Playing Field for Neurodiverse Workers

  • Tech Giants Go All In on AI Debt

Read time: 5 minutes

LATEST DEVELOPMENTS

Evolving AI: Moonshot AI’s new Kimi K2 Thinking model sets a new standard for open-source reasoning and coding.

Key Points:

  • K2 Thinking runs up to 300 tool calls in one session, maintaining logic over hundreds of reasoning steps.

  • With one trillion parameters and a 256K token context, it surpasses major commercial models on reasoning and coding benchmarks.

  • The model generated a full Word-style app from a single prompt and solved PhD-level math problems autonomously.

Details:

Moonshot AI’s Kimi K2 Thinking is a trillion-parameter “thinking agent” designed for multi-step reasoning and agentic tasks. It uses test-time scaling to extend its reasoning depth and performs hundreds of tool calls in sequence without human input. Benchmarks show record scores on reasoning, coding, and agent-based tasks, outperforming some versions of GPT-5 and Claude Sonnet. The model is available on kimi.com, Hugging Face, through the API and the Kimi App.

Why It Matters:

Open-weight models that can think in long chains like this give teams a way to run serious agent workflows on their own infrastructure instead of tying every big research or coding task to closed APIs. K2 Thinking can plan 200–300 tool calls in a row, hit top scores on agentic browsing and tough exams, and still run with INT4 efficiency, which makes deep research loops or repo fixes cheaper to test at scale. For anyone building AI copilots, this means you can route heavy planning, search, and analysis jobs to an open system you can fine tune and self host, then save pricier frontier models for the narrow cases where extra polish or safety tooling really counts.

Simplify Training with AI-Generated Video Guides

Simplify Training with AI-Generated Video Guides

Are you tired of repeating the same instructions to your team? Guidde revolutionizes how you document and share processes with AI-powered how-to videos.

Here’s how:

1️⃣ Instant Creation: Turn complex tasks into stunning step-by-step video guides in seconds.
2️⃣ Fully Automated: Capture workflows with a browser extension that generates visuals, voiceovers, and call-to-actions.
3️⃣ Seamless Sharing: Share or embed guides anywhere effortlessly.

The best part? The browser extension is 100% free.

Source: Teera Konakan / Moment / Getty Images

Evolving AI: Neurodiverse professionals say AI agents are helping them stay focused, organized, and thrive at work.

Key Points:

  • A UK study found neurodiverse workers are 25% more satisfied with AI assistants than neurotypical peers.

  • Tools like meeting summarizers and smart schedulers support executive function and communication.

  • Ethical AI design is key to balancing different needs and avoiding bias or forced disclosure.

Details:

AI assistants are giving neurodiverse employees new ways to succeed. Workers with ADHD, autism, or dyslexia report improved focus and productivity thanks to tools that handle note-taking, summarize meetings, and manage time. Experts say companies investing in AI ethics see stronger ROI and happier teams, though they must address risks like bias and conflicting accessibility needs. The push for more inclusive AI is gaining momentum with projects like Humane Intelligence’s Bias Bounty Challenge.

Why It Matters:

AI at work is quietly turning into assistive tech for brains that were never built for back-to-back meetings and email pings. New studies show neurodivergent employees are both heavier users of AI and more satisfied with it, reporting big gains in focus, communication and access to information compared with neurotypical colleagues. For companies, this is a rare two-for-one: invest in trustworthy, well-governed AI that respects privacy and you get higher ROI plus a workforce where ADHD, autism or dyslexia stops being a hidden tax on performance and starts looking like an edge.

AI DEVELOPMENT RACE
💲 Tech Giants Go All In on AI Debt

Source: Professor Robert J Shiller, Yale University and Bank calculations.

Evolving AI: Big Tech is taking on record debt to fund its massive AI buildout.

Key Points:

  • Google, Meta, Microsoft, and Amazon spent $112B in three months on AI infrastructure.

  • Meta and Blackstone use complex loans and bonds to fund data centers.

  • Banks are backing multi-billion projects like OpenAI’s Stargate facility in New Mexico.

Details:

Major tech companies are borrowing at unprecedented levels to keep up in the AI race. They’re issuing bonds tied to data centers, creating off-book entities to hide debt, and leaning on private lenders for an estimated $800B in fresh capital. OpenAI’s data center expansion alone could top $38B, backed by major banks. The circular nature of tech investment—companies funding AI labs that then buy their cloud services—adds even more financial pressure to the system.

Why It Matters:

Big Tech is turning the AI boom into a giant carry trade: trillions in data centers are being stitched together with SPVs, project loans and asset backed bonds that end up on the books of banks, insurers and pension funds. If AI usage and pricing do not ramp fast enough, that can mean pressure to hike cloud prices, cut generous free tiers or slow product rollouts right as companies bake these tools into daily workflows. For anyone building on OpenAI, Google, Microsoft or Meta, your “cheap” AI stack is effectively subsidised by leverage, which makes vendor choice feel a lot more like a macro bet than a pure tech call.

CTV ads made easy: Black Friday edition

As with any digital ad campaign, the important thing is to reach streaming audiences who will convert. Roku’s self-service Ads Manager stands ready with powerful segmentation and targeting — plus creative upscaling tools that transform existing assets into CTV-ready video ads. Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

 👀 Click on the image you think is real

QUICK HITS

🌊 Underwater cables are a vital piece of the AI buildout and internet.

⛪ Pope Leo XIV urges Catholic technologists to spread the Gospel with AI.

🇪🇺 EU weighs pausing parts of landmark AI act in face of US and big tech pressure.

😟 Social Media Victims Law Center and Tech Justice Law Project lawsuits accuse ChatGPT of emotional manipulation.

👪 What parents need to know about Sora.

📰 ChatGPT’s news picks swing wildly depending on whether you use the web interface or the API.

📈 Trending AI Tools

  • 🚀 Amplemarket - Uncover opportunities and grow revenue with this all-in-one AI platform*

  • 🧠 Kimi K2 Thinking - Moonshot AI’s new open-source thinking agent

  • 🗂️ Tana – A note-taking app built with AI that turns your thoughts into tasks you can track

  • 🎨 Recraft – Generate clean, styled AI images with text support

*partner link

Reply

or to participate.