Edge0 open-sourced a 145B model that runs fully on a laptop at over 80 tokens a second, smolmaxxing got a 16B model running on a TV box and a 2013 iPad with about 1 GB of memory, and Anthropic released Claude Fable 5.2 with faster output and 20% lower API prices. And that's not all, here's everything in AI this week:

Together with Mindstream

Master Claude AI (Free Guide)

The professionals pulling ahead aren't working more. They're using Claude.

Our free guide will show you how to:

  • Configure Claude to be the perfect assistant

  • Master AI-powered content creation

  • Transform complex data into actionable strategies

  • Harness Claude’s full potential

Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work.

Yogurt meeting summary with next steps by person

Frigade released Yogurt, a free, open-source alternative to Granola. It records your microphone and your computer's audio, so no bot joins the call, then transcribes the meeting and writes a summary on your own computer.

  • Next steps by person: The summary merges your own notes with the conversation into a plan, next steps for each person, and your personal reminders. You can also ask questions about a meeting and get answers linked to the transcript.

  • Stays on your computer: Transcription and summaries run locally with Whisper and llama.cpp. After the first model download it works offline, with no account and no telemetry.

  • Free for Mac, Windows, and Linux: It is MIT licensed and needs 8 GB of RAM, and the Mac version needs Apple Silicon. A local MCP server lets you pass notes to Claude, ChatGPT, or other AI apps.

Try it now → https://www.yogurt.so

Cicada: any call, any stream, no language barrier

Cicada is a free, open-source app that listens to whatever plays on your computer and gives you live bilingual subtitles, real-time translation, and structured meeting notes. It now runs on Mac, Windows, Linux, and Android.

  • Any call or stream: It reads system audio, so it works with Zoom, Teams, Google Meet, Discord, X Spaces, YouTube, and more, with no bot joining the call.

  • Speak back in another language: Hold a key (right Option on Mac, right Ctrl on Windows) and talk, and the other people in the meeting hear a translated version of what you said.

  • Free, with one catch: On Apple Silicon Macs, speech recognition runs on your computer. Windows, Linux, Android, and Intel Macs use a cloud service for it, so they need your own AI key (Ling, OpenRouter, or ZenMux). The license is Apache 2.0.

n8n Nod handling email, calendar and a dinner booking in chat

n8n Labs released Nod on November 10 as a research preview. It is a free personal agent that runs on your computer in its own sandbox, with its own browser, and finishes jobs across your apps and websites.

  • Works across your apps: It connects to Gmail, Calendar, Slack, Notion, and Drive, and uses its own browser for sites with no integration, for jobs like cancelling subscriptions, tracking prices, or booking hotels.

  • Asks before it acts: It checks with you before sensitive steps, keeps saved passwords and cards out of the model's view, shows what each task cost, and lets you set a monthly spending limit.

  • Free, bring your own model: The app is Apache 2.0 and needs no n8n account. You pay your AI provider or run a local model. It runs on macOS 14 or later, with Windows 11 and Linux in preview.

Try it now → https://nod.n8n.io

Claude Fable

Anthropic released Claude Fable 5.2 on November 9, alongside Claude Mythos 5.2. It is the most capable model Anthropic offers to the public, with improvements in coding, reasoning, and vision. Mythos 5.2 uses the same underlying model but is limited to vetted organizations.

  • Faster and cheaper: Outputs come up to 30% faster than Fable 5.1, and the API price drops 20% to $8 per million input tokens and $40 per million output tokens. A fast mode speeds up generation by up to 2.5x.

  • Fewer blocks: It keeps its cybersecurity and biology safeguards, but blocks about 55% fewer requests than Fable 5.1. The context window is 1 million tokens.

  • Where to get it: It is on Pro, Max, Ultra, Team, and Enterprise plans, the Claude Platform, AWS, Google Cloud, and Microsoft Foundry. Anthropic says a team of 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in 30 hours.

Try it now → https://claude.ai

smolmaxxing chat page streaming tokens from a model running on a TV box

smolmaxxing is an open-source project that runs the biggest language model it can on the smallest hardware. It got a 16 billion parameter model running on a Xiaomi TV box and a 2013 iPad Air, each with about 1 GB of usable memory.

  • How it fits: The 16B model, Ling-mini-2.0, is a mixture-of-experts model. The most used experts stay in RAM and the rest stream from flash storage, with custom kernels for each chip.

  • Real speeds: It runs at 5.1 tokens a second on the TV box and 2.6 to 4.5 on the iPad. A 2B model runs at 7.7 tokens a second on the TV box, and a 1-bit 1.7B model reaches 24 to 26 on the iPad.

  • Even bigger on a phone: A 2022 OnePlus Nord CE 2 runs the 80B Qwen3-Next model with about 3 GB free. It is an MIT-licensed patch for llama.cpp, and the iPad has to be jailbroken.

Edge0 logo

Edge0 released Edge0-145B on November 12, a 145 billion parameter model built to run fully on a laptop. Only about 10 billion parameters are active at each step, so it peaks at 29 GB of memory, and Edge0 reports more than 80 tokens a second.

  • Laptop numbers: Edge0 reports 1,100+ tokens a second for reading prompts, 80+ for writing, about 120 W of power, and a 128K to 256K context window. These are the company's own figures, and it does not say which laptop it used.

  • Strong scores: Edge0 reports 86.52 on Terminal-Bench 2.1, ahead of Qwen3.8-Flash-Next and DeepSeek-V4-Flash, and 90.4 on GPQA-Diamond.

  • What you get: Weights for Edge0-145B and a smaller Edge0-120B, plus the inference engine, for macOS, Windows, and Linux on Apple M-series chips and NVIDIA RTX GPUs. The weights use Edge0's own community license, and the engine is Apache 2.0.

Thanks for making it to the end! I put my heart into every email I send. I hope you are enjoying it. Let me know your thoughts so I can make the next one even better.

See you tomorrow :)

Dr. Alvaro Cintas

✓ Full archive of premium guides with ready-to-use prompts

✓ Structured AI courses (step-by-step, start-to-finish)

✓ Every upcoming premium tutorial

✓ $500+ worth of AI tools