This Week In OS AI Field Notes: 20 November 2026

This week’s open-source AI updates: new models from IFM, Thinking Machines and NVIDIA, agent tools, policy, podcasts, and the latest from Sentient.

This Week In OS AI Field Notes: 20 November 2026

Open-Source Models

Open-Source Models — OS AI Field Notes, 20 November 2026

  • PAN 2: Global AI research lab Institute of Foundation Models released PAN 2, an open physical intelligence model that lets AI agents and people see what would happen before they act. It simulates outcomes that follow real physics, respects the timing of each process, and keeps the scenery and cuts consistent as the camera moves. [ref] [10 NOV 2026]

  • INKLING 2 SMALL: Thinking Machines released the open weights for Inkling 2 Small, a general-purpose multimodal model that accepts text and image inputs and generates text outputs. It is now the #1 US open model and #2 among all open models on Agent Arena. It’s also the most steerable open model on the board, with a median cost of $0.26 per task, about 4.7x cheaper than Kimi K3.1 Pro. [huggingface] [ref] [12 NOV 2026]

  • QWEN4 FAMILY: Alibaba announced that open weights for smaller Qwen4 models are on the way, alongside the launch of its closed-weight flagship, Qwen4-Max. Built on a new MoE architecture with native vision and a 1M-token context window, Qwen4-Max outperforms Qwen3.8-Max with up to ~50% fewer tokens and cuts average cost per task by 62%. [ref] [15 NOV 2026]

  • NEMOTRON-LABS-SIGMA: NVIDIA released open weights for Sigma, the first large-scale continuous diffusion language model. Most models write text one word at a time. Sigma manipulates and refines blocks of text all at once, which makes its output easier to steer and faster to generate. On math and coding, Sigma-3B outperforms the open diffusion models LLaDA-8B and Dream-7B—half their size and 3.5 times faster. [paper] [huggingface] [16 NOV 2026]

  • LIQUID AI D1.2: Liquid AI released d1.2 and d1.2-ultra, decision models that take text, images, and audio and return structured answers. They can pull a value from an invoice, find every piece of personal data in a note, or locate a signature on a scanned form. d1.2-ultra ranks #1 on the Artificial Analysis Decision Index and is the best at knowing when it’s unsure, beating GPT-6 Luna and Perplexity’s Decider. Pricing starts at $0.02 per million input tokens, with no charge for output. [ref] [16 NOV 2026]

Tools & Frameworks

Tools & Frameworks — OS AI Field Notes, 20 November 2026

  • NOD: n8n Labs released Nod, a standalone open-source personal agent free under the Apache 2.0 license. It runs on your own computer with a sandboxed virtual machine and browser, works inside connected apps and on websites, asks before taking actions, and can use any model provider you bring, including local models. [ref] [github] [11 NOV 2026]

  • HUGGING FACE & OPENROUTER: Hugging Face and OpenRouter partnered to make open models easier to use, with the former handling storage and the latter inference. Hugging Face’s Inference Providers now route language model requests through OpenRouter’s 80+ providers, choosing the best mix of price, speed, and reliability per call. OpenRouter users can also access Hugging Face’s open weights and store agent traces there to inspect, share, or train on. [ref] [16 NOV 2026]

Policy & Security

Policy & Security — OS AI Field Notes, 20 November 2026

  • EUROPA: The European Commission is building EUROPA, the EU’s open frontier model for public authorities, researchers, and businesses across all 24 official EU languages. Domyn, the Milan company chosen to develop the model, ran training trials on JUPITER, Europe’s first exascale supercomputer, which made training 14 times faster. Full training is expected to take about four months. [ref] [16 NOV 2026]

  • CLOUD COMPUTING RESTRICTIONS: Trump administration moves to block China from indirect access to American AI tech. The U.S. Department of Commerce issued an immediate rule restricting remote access to advanced U.S. AI chips and servers. Overseas data centers serving countries like China and Russia must now identify end users, while chip exports to high-diversion-risk countries such as Malaysia, Singapore, Indonesia, Thailand, and Brazil face additional licensing requirements. [ref] [16 NOV 2026]

Podcasts & Media

Podcasts & Media — OS AI Field Notes, 20 November 2026

  • ALL-IN PODCAST’S JENSEN OUT: Jason Calacanis, Chamath Palihapitiya, David Friedberg, and David Sacks discuss why Anthropic’s race toward recursive self-improvement to build a monopoly is “the most dangerous thing” in AI, and how Jensen Huang’s exit as Nvidia CEO affects the open-source movement. [youtube] [13 NOV 2026]

  • THE PRAGMATIC ENGINEER WITH SATYA NADELLA: Gergely Orosz talks with Microsoft CEO Satya Nadella about why every company should own its AI stack instead of relying on a single lab. Nadella sees open-weight models as a way for firms to keep control of their knowledge, pushes for swappable models and local “unmetered intelligence” on laptops, and explains how software engineering is changing as code becomes cheap. [youtube] [16 NOV 2026]

  • OPEN COMMONS PODCAST, DON'T LET AI SPEND YOUR MONEY WITHOUT PERMISSION: Coinbase Developer Platform’s Kevin Leffew joins host Sachi Kamiya to discuss x402, the open standard that revives HTTP’s 402 “Payment Required” code from 1997 so AI agents can pay for data and APIs on the spot. He explains why Coinbase handed x402 to the Linux Foundation and how AI is breaking the ad-funded web. [youtube] [17 NOV 2026]

What’s New at Sentient

  • JULIE BELIÃO JOINS SENTIENT FOUNDATION: Julie Belião, a Google veteran who previously led product development at Mozilla.ai, has joined the Sentient Foundation as VP of Product. With more than 15 years in AI, she’ll lead product strategy through Sentient Labs to turn the team’s open-source AI research into tools developers and organizations can use. [ref] [10 NOV 2026]

  • SENT LISTED ON ROBINHOOD LEGEND: SENT is now available to trade on Robinhood Legend, Robinhood’s free, browser-based desktop platform with advanced charting and analysis tools. [ref] [12 NOV 2026]

Sentient Foundation is a non-profit organization dedicated to developing open-source AGI, free from single-entity control. We accelerate AGI research and development through Sentient Labs and a global community of open-source builders.

We're building intelligence that dreams. Dreaming machines, designed in the open. Advancing our shared future through open-source AGI.

Visit Sentient Foundation website, follow us on LinkedIn and Twitter/X.

Subscribe to open source AI news
with OS AI Field Notes