🔥 OpenAI Prepares Astra for Launch

PLUS: Nvidia Buys Hugging Face

Author

The Summary AI
August 27, 2026

Sponsored by

Welcome back!

OpenAI is preparing to launch its next frontier model, Astra, built to run many AI agents at once and tackle long, complex tasks on its own. But its growing skills are also triggering OpenAI’s highest cyber risk level, forcing the company to tighten security and possibly delay the release. Let’s unpack…

Today’s Summary:

  • 🔥 OpenAI prepares Astra

  • 🎥 Gemini Omni 1.1 Video

  • đź’» Qwen3.8 brings frontier AI locally

  • 🤗 Nvidia buys Hugging Face

  • ⚡ GLM-5.3 Flash cuts frontier costs

  • 🚀 Gemini 3.7 Flash improves coding

  • 🛠️ 2 new tools

TOP STORY

OpenAI Astra nears critical cyber capability

The Summary: OpenAI is preparing its next frontier model, Astra, meant to run several AI agents at once and operate computers on its own. Early tests show it was able to solve decade-old math problems and perform sophisticated research work. Those same capabilities raised a red flag: OpenAI says it can't rule out Astra having its highest-risk cyber capability level, leading the company to pause some frontier work while stronger security controls are put in place.

Key details:

Why it matters: Every major lab now faces a problem: agent skill and cyberattack skill seem to grow on the same curve, so you can't ship one without shipping the other. With Astra, the real product may be the containment system built to hold the intelligence.

FROM OUR PARTNERS

Your AI Agent Follows You Everywhere

AI agents built to get work done

Skydive agents take on work for you, using the same tools your team already uses.

Talk to your agent on the web, Slack, email, iMessage, or right from your terminal. Wherever you pick up the conversation, your agent keeps the context.

Put agents to work across customer support, sales, marketing, engineering, ops, and more.

Give your agent a job. They’ll take it from there.

GOOGLE

Google launches Gemini Omni 1.1 Flash Video

The Summary: Google’s Gemini Omni 1.1 Flash gives more control over AI video, from extending existing scenes to choosing exact first and last frames. The model can study 10 seconds of prior footage, preserve characters and lighting, and extend a video up to 40 seconds. A new draft mode cuts generation cost to 1/3 while running 60% faster. Finished clips can be up to 4K.

Key details:

  • Scene extension reads 10 seconds of prior footage and adds footage in 10-second blocks up to 40 seconds total

  • Users can feed in up to three seconds of a reference video so the model can understand its motion patterns, visual context, and characters

  • First-and-last-frame controls hand the model two keyframes and ask it to create the motion between them, for dolly zooms, camera orbits, transitions, and seamless loops

  • Rolling out in Google Flow, Google AI Studio, and Gemini Agent Platform

Why it matters: AI video is starting to look like an editing tool with many control knobs. Reference footage and long scene memory attack one of generative video’s hardest problems, keeping visual logic coherent from shot to shot.

FROM OUR PARTNERS

22 AI Agents Ready in 5 Minutes

22 ChatGPT Agents Built for Every Marketing Job

Most marketers use ChatGPT to do general research and then call it an AI strategy. The ones outperforming them are deploying specialized agents built for specific jobs.

We put together 22 plug-and-play ChatGPT marketing agents that handle the work eating your week, each with built-in instructions and structured outputs ready to go in under 5 minutes.

Subscribe to Marketing Against the Grain and get all 22 free.

Inside you'll find:

  • Competitive intelligence agent that visits competitor websites and builds detailed comparison matrices automatically

  • Customer feedback analyzer that ranks improvement opportunities by business impact

  • Social listening specialist that monitors brand mentions and flags reputation risks before they escalate

  • Campaign optimization agents that handle attribution analysis and surface what is actually driving results

Your competitors are already running agents like these.

Get 22 ChatGPT Marketing Agents free when you subscribe to Marketing Against the Grain today.

QWEN

Qwen3.8 new model fits in a 17GB file

The Summary: Alibaba released Qwen3.8-27B, an open-weight model small enough to run on a laptop but strong enough to perform near frontier level. The model ships under the free Apache 2.0 license, supports context up to 262K tokens and understands images and video.

Key details:

  • #9 overall on Code Arena, the only model of its size in the top 10

  • Qwen says it beats Opus 4.6 Max on several practical tests: 61.7% vs. 53.4% on software-engineering tasks, and 84.3% vs. 72.7% on computer-use tasks

  • The incredible part is that all this fits in a 17GB file. You can run it on a MacBook and it can code, use tools, inspect images and work with long context

  • The tradeoff is speed. Local runs reach about 15-30 tokens per second, while paid frontier models respond much faster, especially on complex jobs

Why it matters: Qwen3.8-27B makes a useful new setup possible: let the local free model do the long coding grind, and use a stronger paid model for the planning and final review. That can cut API costs a lot. In many cases, the paid frontier model may become an escalation model instead of the default.

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/