Plus: AWS's 3nm chip rewriting the training playbook, Mistral's open-weight challenger to GPT-5, and 1,000 Amazon workers sounding the alarm
Sponsored by
Create your AI voice clone from just a few minutes of audio with Eleven Labs. No more need to record multiple takes to get the audio for your content right. The best thing? Your clone can even speak almost 30 languages!
Give it a tryThis week in AI felt like a turning point. OpenAI issued an internal "code red" as Google and Anthropic close the gap, Mistral dropped an open-weight model claiming parity with GPT-5, and AWS unveiled a 3nm chip that could reshape how we train frontier models.
Oh, and over 1,000 Amazon employees signed a letter warning that the AI race is moving too fast for anyone's good.
If you've been paying attention to this space, you know the major players rarely show their cards. This week? Everyone's hand is on the table.
On this day in 2024, Google announced Gemini 2.0 and Veo 2, their breakthrough video generation model. One year later, Runway just claimed the top spot on the Video Arena leaderboard, beating both Google and OpenAI at their own game.
The competition is real. The stakes keep climbing. And the pace isn't slowing down.
Let's dive into what happened this week.
🚨 OpenAI Issues Internal "Code Red" Amid Fierce Competition
Sam Altman sent a memo to staff on December 2 outlining urgent efforts to improve ChatGPT as Google and Anthropic gain serious ground. Despite 800M weekly users, OpenAI is delaying advertising products and shopping features to focus on core functionality. The company is feeling real pressure for the first time since launching ChatGPT.
🤖 Mistral Drops Mistral Large 3: "World's Best Open-Weight Multimodal Beast"
Mistral Drops Mistral Large 3: "World's Best Open-Weight Multimodal Beast" Unveiled December 2, this 124B-param monster crushes in multilingual reasoning, vision, and robotics—claiming parity with GPT-5 and Gemini 3 Pro while running on-device. Paired with Ministral minis for drones and agents, it's Europe's shot at US AI supremacy.
🚨AWS Drops Trainium3: A 3nm AI Chip Powering 4x Faster Training at re:Invent 2025
At AWS re:Invent on December 2, Amazon unveiled the Trainium3 UltraServer—a breakthrough 3nm AI chip delivering up to 4.4x more compute and 4x better energy efficiency than Trainium2. Scalable to massive clusters and fully compatible with Nvidia hardware, it's set to cut costs for training frontier models and accelerate agentic AI for devs everywhere.
🎯 Over 1,000 Amazon Employees Warn AI Rollout Risks Democracy, Jobs, and Climate Goals
In an open letter dated November 29, Amazon workers from cashiers to IT pros slammed the company's aggressive AI push, accusing it of ditching climate targets, automating jobs, and enabling surveillance states amid authoritarian rises—calling for ethical pauses in the tech's expansion.
💻 Runway's Gen 4.5 Tops Video Arena Leaderboard, Outshining Google and OpenAI
Runway unveiled Gen 4.5, a groundbreaking AI video model that generates high-definition clips from text prompts and claims the top spot on the Video Arena benchmark—surpassing rivals like Google's Veo and OpenAI's Sora. Rolling out to all users this week via API and partnerships, it's a boon for creators building immersive tools and agents.
⚡ Cline topped GitHub's Octoverse 2025 as the fastest-growing AI project by contributors, with 4,704% YoY growth. This autonomous coding agent lives in your IDE and handles complex development tasks step-by-step with your approval. It even supports Model Context Protocol (MCP) to extend its own capabilities.
🔧 Google Cloud Agent Starter Pack provides production-ready templates for building GenAI agents on Google Cloud. With pre-built ReAct and RAG patterns, integrated Vertex AI evaluation, and one-command CI/CD setup, it's everything you need to ship agents fast without reinventing infrastructure.
🚀 vLLM continues dominating as the go-to inference engine for LLMs. With PagedAttention delivering up to 24x higher throughput than HuggingFace Transformers, it's become essential infrastructure for anyone deploying models at scale. Six of the 10 fastest-growing open source repos were AI infrastructure projects like this.
🔗 Model Context Protocol by Anthropic is becoming the "USB-C for AI applications." This open standard for connecting AI systems to data sources has been adopted by OpenAI, Google DeepMind, Replit, Zed, and Sourcegraph. Pre-built MCP servers exist for Google Drive, Slack, GitHub, Postgres, and more.
Coming in Hot
Create professional AI videos instantly.
AI meeting assistant that turns conversations into work.
Make research papers easy to understand and connect.
Instant AI answer generator & problem solver.
Smarter research with AI that adapts to you.
Simplify content creation with AI-powered blogging.
Your AI concierge that handles calls and keeps you focused on creative work.
Effortless translation of webpages, documents, images and videos in one tool.
Search and organize files across 20+ apps with AI-powered semantic search.
Type a keyword, get your full prompt instantly across any AI platform.
Submit your AI tool to the #1 AI tools marketplace.
This week we're spotlighting three tools that caught our attention for different reasons—speed, data extraction, and open-source coding power.
What it does: A multimodal AI acceleration platform that supercharges image and video generation/editing from text, images, or audio inputs, integrating top models for rapid, high-quality content creation without the hassle of managing infrastructure.
Key features:
Pricing:
Recent updates: Live demos at AWS re:Invent 2025 (Dec 1 to 5), Kling Image O1 Update 2 released Dec 2, new ParaAttention caching papers, and continuous model rollouts throughout 2025.
Why it matters: Speed wins in content creation. WaveSpeedAI gives solo creators and agencies production-grade visuals at prototype velocity without managing GPUs or fighting queue times.
What it does: AI-powered no-code web extraction tool that captures, structures, and encrypts data from single pages or entire sites, then delivers it directly to Google Sheets or CSV.
Key features:
Pricing:
Recent updates: November 2025 competitive pricing scraper launch, new group research collaboration features, expanded e-commerce templates, and enterprise encryption certifications.
Why it matters: Most scraping tools force you to choose between ease of use and security. Pline delivers both, letting non-technical teams pull clean, compliant data that actually drives decisions.
What it does: Kilo Code is an open-source agentic coding assistant for VS Code and JetBrains that emphasizes privacy, model flexibility (cloud or local), and project memory. It's become one of the most talked-about AI development tools of 2025.
Key features:
Pricing:
Recent updates: Spectre model launch Dec 1 2025 with 256k context, MiniMax M2 integration surge, JetBrains Marketplace approval, CLI v4 releases, and Advent of Code 2025 sponsorship with live agent demos.
Why it matters: Closed tools like Cursor and Copilot own your context and lock your costs. Kilo Code returns power to developers with true model choice, verifiable privacy, and agentic workflows that already power over 500k daily users.
🛡️ Trump Administration Targets State AI Laws, Then Backs Down
The Trump administration drafted an executive order to challenge state AI regulations through lawsuits and threatened to withhold federal broadband funding from states with contested AI laws. But after the Senate stripped a federal preemption provision from the One Big Beautiful Bill in a 99-1 vote, the executive order was put on hold. Meanwhile, 38 states have already adopted over 100 AI-related laws in 2025.
🧬 International AI Safety Report Shows Progress and Gaps
The number of companies publishing Frontier AI Safety Frameworks has more than doubled since the 2025 report's initial publication. However, researchers warn that sophisticated attackers can often bypass current defenses, and the real-world effectiveness of many safeguards remains uncertain. The report highlights that new training techniques teaching AI systems step-by-step reasoning have driven recent capability advances—with implications for biological weapons and cyber attack risks.
💼 Microsoft 365 Copilot Business Launches at $21/Month
Starting December 2025, Microsoft 365 Copilot Business becomes available for SMBs with under 300 users at $21/user/month—making AI-powered productivity tools more accessible to smaller teams. The company also launched a Sales Development Agent that autonomously builds pipeline, nurtures leads, and personalizes outreach.
🎤 ChatGPT Gains Image Library and Adaptive Reasoning
OpenAI rolled out GPT-5.1 Instant and GPT-5.1 Thinking models on November 24. For the first time, GPT-5.1 Instant can use adaptive reasoning to decide when to think deeply before responding. All images created with ChatGPT now auto-save to a new Library feature. Pro, Enterprise, and Edu users also got ChatGPT Record for transcribing and summarizing meetings.
🇦🇺 Australia Launches National AI Plan And New Government AI Policy
Australia released a National AI Plan that aims to accelerate AI adoption across the economy while mostly relying on existing laws rather than creating heavy new AI‑specific regulation. In parallel, the federal government overhauled its internal AI policy with a new impact‑assessment tool and procurement guidance for “responsible AI” in public services, effective mid‑December.
🥊 Windsurf vs Cursor: The AI IDE Battle Intensifies
Windsurf has emerged as Cursor's main competitor with its Cascade agent feature, proprietary SWE-1.5 model (13x faster than Sonnet 4.5), and cleaner UI. Starting at $15/month compared to Cursor's $20/month, Windsurf targets large codebases with deeper multi-file understanding. Both tools are shipping weekly updates as the AI coding assistant market heats up.
We pulled together the best Black Friday deals from AI tools in the directory.
Heads up, though: the deals page comes down this week, and a bunch of these discounts vanish after Cyber Monday. If there's a tool you've been eyeing, don't sleep on it.
