> ## Content Index
> Fetch the complete content index at: https://bullquill.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI Weekly #1: Gemini 4 Argon, Meta's Open Muse Spark, EU AI Act
- URL: https://bullquill.com/ai-weekly-1-gemini-4-argon-muse-spark-open-eu-ai-act/
- Published: 2026-10-20T08:00:00.000Z
- Updated: 2026-10-21T12:44:33.000Z
- Description: Gemini 4 Argon arrives, Meta opens Muse Spark weights, DeepSeek updates Flash, and the EU publishes its first AI Act standard. What it means.
- Author: Rowan Ashby
- Tags: AI, ai-weekly, Gemini, Meta, DeepSeek, Anthropic, eu-ai-act, pytorch, open-weight-models, #lang-en

*By Rowan Ashby · October 20, 2026 · Covers October 13–20, 2026*

Welcome to the first edition of AI Weekly from Bullquill. Each week we pick the AI changes that matter most, explain them in plain language, and link to the original announcement so you can check it yourself. This week's theme is access: big models became easier to use, through a stable API or as downloadable weights, and Europe published the first technical standard tied to its AI law.

---

## 1\. Google makes Gemini 4 Argon generally available (October 13)

Google's Gemini API changelog lists Gemini 4 Argon (`gemini-4-argon`) as generally available from October 13 and calls it "our most capable model built for complex coding, agentic, math, and science tasks." The same day, Google said Google AI Pro and Ultra subscribers can use it in the Gemini app. General availability means a stable model ID that businesses can build on, not a preview that may change or vanish. For prices, Google points developers to its "What's new in Gemini 4 Argon" page.

**Sources:** [Gemini API changelog](https://ai.google.dev/gemini-api/docs/changelog?ref=bullquill.com) · [Google: Gemini 4 Argon is now available in the Gemini app](https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/?ref=bullquill.com)

## 2\. Meta releases Muse Spark Open weights (October 15)

Meta published Muse Spark Open on Hugging Face, a mixture-of-experts model with 816 billion total parameters, 42 billion of them active per token, and a context window of up to 1 million tokens. The weights cost nothing, but access is gated: you submit a request form on Hugging Face and wait for manual approval before you can download. The model also ships under Meta's own Muse Model License, which requires products to display "Built with Muse Spark Open," and companies offering it as an API-style service with combined gross revenue above US$20 million in any 12-month period must notify Meta and sign a separate agreement. (Our guide explains how "open-weight" differs from "open-source.")

**Sources:** [Muse Spark Open model card and license, Hugging Face](https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B?ref=bullquill.com) · [Hugging Face launch post](https://huggingface.co/blog/meta-muse-spark-open?ref=bullquill.com)

## 3\. DeepSeek updates V4.1 Flash and opens a Search & Fetch API (October 17)

DeepSeek's changelog lists an upgrade to DeepSeek-V4.1-Flash-1017, which also replaces the model in its web and mobile apps; developers calling `deepseek-flash` get it automatically, at unchanged prices. The weights are on Hugging Face under the permissive MIT license with no access gate, and the model card calls it a minor update with better reasoning, coding and agent scores (DeepSeek's own benchmarks) and shorter, cheaper outputs. The same day, DeepSeek opened API access to the search tools behind its own products, so small teams can get a model and web search from one provider.

**Sources:** [DeepSeek API change log](https://api-docs.deepseek.com/updates?ref=bullquill.com) · [DeepSeek-V4.1-Flash-1017 model card](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017?ref=bullquill.com)

## 4\. Anthropic sets up a Public Benefit Fund backed by its own shares (October 18)

Subject to stockholder approval, Anthropic will commit 5% of its shares to a new Anthropic Public Benefit Fund now. The board may then add another 2% each time the company's value grows 50% beyond a baseline it has not yet disclosed. The fund is a wholly owned LLC rather than a charity, which Anthropic says gives it flexibility in how it gives, and the company states: "None of the Anthropic equity held by the LLC will be used for lobbying or political giving." Governance details are promised "in the coming months," so the fund's independence and eventual size remain open questions.

**Source:** [Anthropic: Investing in the public benefit](https://www.anthropic.com/news/public-benefit-fund?ref=bullquill.com)

## 5\. The EU publishes the first harmonized standard under the AI Act (October 19)

The European Commission published in the Official Journal a reference to EN 18286:2026, a CEN and CENELEC standard describing a quality management system for providers of high-risk AI. Under Commission Implementing Decision (EU) 2026/2274, providers that apply it are presumed to meet the parts of Article 17, the AI Act's quality-management duty, that the standard addresses. That turns a broad legal obligation into a concrete checklist auditors can recognize, which usually lowers compliance cost. The Commission says more harmonized standards will follow before the high-risk rules apply.

**Source:** [European Commission: First harmonised standard supporting the AI Act published in the Official Journal](https://digital-strategy.ec.europa.eu/en/library/first-harmonised-standard-supporting-ai-act-published-official-journal?ref=bullquill.com)

## 6\. AWS brings Trainium chips into PyTorch as a native device (October 19–20)

AWS announced TorchNeuron, an open-source (Apache 2.0) PyTorch backend that lets developers move a model onto AWS Trainium chips as the `neuron` device, much as they would with a GPU, in both eager mode and `torch.compile`; the public beta starts October 20\. The fewer code changes needed to switch hardware, the more competition Nvidia GPUs face, which in time pushes down the cost of running AI. The announcement says code written against PyTorch's device-agnostic API "runs unmodified," but its own test results list unsupported operators, so not every model will work on day one.

**Source:** [PyTorch blog: Introducing TorchNeuron](https://pytorch.org/blog/introducing-torchneuron-native-pytorch-backend-for-aws-trainium/?ref=bullquill.com)

---

## What it means for you

- **App users:** DeepSeek's app now runs the updated Flash model, and Gemini 4 Argon is in the Gemini app for Google AI Pro and Ultra subscribers.
- **Builders:** two large models became downloadable this week under very different terms. Read the license, and check for access gates, before you ship.
- **EU-regulated businesses:** EN 18286 is the first concrete standard to map your AI processes against.

## What to watch next week

- Google's changelog says the `deep-research-pro-preview-12-2025` agent in the Gemini API shuts down on **October 23, 2026**. ([Source](https://ai.google.dev/gemini-api/docs/changelog?ref=bullquill.com))
- Independent benchmarks for Muse Spark Open and DeepSeek-V4.1-Flash-1017, which so far rest on the companies' own numbers.

## Sources

All accessed October 20, 2026.

1. Google, Gemini API changelog: [https://ai.google.dev/gemini-api/docs/changelog](https://ai.google.dev/gemini-api/docs/changelog?ref=bullquill.com)
2. Google, "Gemini 4 Argon is now available in the Gemini app" (October 13, 2026): [https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/](https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/?ref=bullquill.com)
3. Meta, Muse Spark Open model card and Muse Model License: [https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B](https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B?ref=bullquill.com)
4. Hugging Face, "Welcome Muse Spark Open" (October 15, 2026): [https://huggingface.co/blog/meta-muse-spark-open](https://huggingface.co/blog/meta-muse-spark-open?ref=bullquill.com)
5. DeepSeek, API change log (October 17, 2026): [https://api-docs.deepseek.com/updates](https://api-docs.deepseek.com/updates?ref=bullquill.com)
6. DeepSeek, DeepSeek-V4.1-Flash-1017 model card: [https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017?ref=bullquill.com)
7. Anthropic, "Investing in the public benefit" (October 18, 2026): [https://www.anthropic.com/news/public-benefit-fund](https://www.anthropic.com/news/public-benefit-fund?ref=bullquill.com)
8. European Commission, "First harmonised standard supporting the AI Act published in the Official Journal" (October 19, 2026): [https://digital-strategy.ec.europa.eu/en/library/first-harmonised-standard-supporting-ai-act-published-official-journal](https://digital-strategy.ec.europa.eu/en/library/first-harmonised-standard-supporting-ai-act-published-official-journal?ref=bullquill.com)
9. PyTorch blog, "Introducing TorchNeuron: Native PyTorch backend for AWS Trainium": [https://pytorch.org/blog/introducing-torchneuron-native-pytorch-backend-for-aws-trainium/](https://pytorch.org/blog/introducing-torchneuron-native-pytorch-backend-for-aws-trainium/?ref=bullquill.com)