Meta and Google Release New AI Models as Industry Launches Accelerate

Meta and Google Release New AI Models as Industry Launches Accelerate In early September, Meta and Google launched new AI models aiming to balance frontier capabilities with lower costs, while reports surfaced regarding OpenAI's reasoning and safety monitoring techniques for its upcoming model. Meta released Muse Spark 1.3 (Max), scoring 62 on the Artificial Analysis Intelligence Index, and plans to release its larger 'Watermelon' model and open weights next. Google introduced Gemini 3.8 Flash, maintaining previous pricing while improving coding, reasoning, and agentic task benchmarks to score 59 on the index. OpenAI's upcoming 'Astra' model reportedly uses 'recurrent depth' looping techniques to boost performance, prompting safety questions over how internal reasoning is monitored. The U.S. government filed a legal brief supporting OpenAI in its copyright dispute with The New York Times, citing national tech leadership. New York City public schools announced a ban on AI usage for students until the eighth grade. Model Launches by Meta and Google Meta launched Muse Spark 1.3 (Max), which scored 62 on the Artificial Analysis Intelligence Index, placing it just behind Claude Fable 5.1 and Opus 5 at lower operational costs. Mark Zuckerberg noted that weights will be released and a larger model codenamed 'Watermelon' is planned next. Meanwhile, Google released Gemini 3.8 Flash at unchanged pricing ($0.75 / $3.75), reaching a 59 index score with improvements in coding and reasoning, though DeepMind's Koray Kavukcuoglu acknowledged it remains slightly below the frontier. OpenAI 'Astra' Safety and Recurrent Loops OpenAI is preparing to launch a model called 'Astra' that reportedly uses a technique known as 'recurrent depth,' analyzing text in iterative loops to extract more intelligence without increasing model scale. While this technique boosts coding and computer-use capabilities, it generates mathematical internal thinking that can obscure reasoning from standard safety monitoring. OpenAI adjusted the model to retain written reasoning outputs and committed to additional launch monitoring, though Chief Scientist Jakub Pachocki warned about the fragility of current monitoring approaches. Industry and Policy Developments U.S. Commerce Secretary Howard Lutnick stated that the administration trusts Anthropic following improved relations. In ongoing litigation between The New York Times and OpenAI, the U.S. government filed an amicus brief arguing that restricting fair use for AI training could jeopardize the country's global technological standing. At the municipal level, New York City public schools instituted a ban on AI tools through eighth grade under Mayor Zohran Mamdani. In Europe, Spanish startup Multiverse Computing launched Quasar 438B, achieving a 43 on the Intelligence Index.

Meta与谷歌发布新模型，OpenAI新技术引发安全监控担忧 9月初AI模型发布潮持续，Meta与谷歌相继推出新模型；同时，OpenAI即将发布的新技术引发安全监控讨论，美政府在版权诉讼与监管立场上也出现新动向。 Meta推出Muse Spark 1.3并在AA智能指数获62分，谷歌推出Gemini 3.8 Flash并获59分 OpenAI新模型Astra采用循环深度技术提升能力，但纯数学推理过程给安全监控带来挑战 美政府在《纽约时报》诉OpenAI案中提交辩护状，支持AI训练的合理使用 纽约市公立学校计划在八年级前禁止使用AI，西班牙初创公司发布Quasar 438B模型 Meta与谷歌发布新模型竞争加剧 Meta推出Muse Spark 1.3，以较低成本在AA智能指数拿下62分，仅次于Claude Fable 5.1与Opus 5；扎克伯格透露下一步将推出代号为“Watermelon”的更大模型并开源Spark权重。谷歌发布Gemini 3.8 Flash，维持原价并提升代码与代理任务表现，智能指数获59分，DeepMind高管坦言当前仍略低于最前沿水平。 OpenAI新架构循环深度引安全监控争议 OpenAI预计推出的Astra模型采用“循环深度”技术，通过循环分析文本榨取智能，但其纯数学思考输出降低了可读性。OpenAI据报调低了循环程度以保留文字推理，首席科学家Jakub Pachocki表示希望避免陷入不可监控的竞争，但也承认当前监控体系本身较为脆弱。 行业快讯与监管动态 美国商务部长Howard Lutnick称政府对Anthropic表示信任。美国政府在《纽约时报》诉OpenAI版权案中递交辩护状，认为限制AI训练的合理使用将损害美国技术领先优势。此外，纽约市公立学校拟在八年级前禁用AI，西班牙初创公司Multiverse Computing则推出了欧洲领先的开源模型Quasar 438B。

[Read Online](https://therundownai.beehiiv.com/p/meta-google-join-the-ai-launch-party) ^| [Sign Up](https://www.therundown.ai/subscribe?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) ^| [Advertise](https://therundownai.typeform.com/to/kraZ1TSO?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) **Good morning, {{ first\_name | AI enthusiasts }}, and welcome to our 3,428 new readers.** We’re only three days into September, and a flurry of new AI models has already hit the leaderboards. Both Meta and Google just joined the launch party Anthropic kicked off, but on very different trajectories — Zuck and co. keep climbing into rarefied AI territory, while Google is out to prove it still has a pulse after a year to forget. **In today’s AI rundown:** Meta, Google join the September AI launch party Nate's Notebook: Tech literacy = your AI ceiling Nail job interviews with the “Proof Project” method Report: OpenAI’s loops hit AI safety monitoring **LATEST DEVELOPMENTS** Image source: Meta/ Google **The Rundown:** Meta and Google both just [shipped](https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) new models, with Mark Zuckerberg pitching Meta’s Muse Spark 1.3 as “frontier performance almost too cheap to meter,” and Google’s Gemini 3.8 Flash showing a stronger bounce-back from its 2026 struggles. **The details:** Spark 1.3 (Max) comes in at a 62 on AA’s [Intelligence Index](https://x.com/ArtificialAnlys/status/2095247787277553929?s=20&utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001), now trailing only Claude Fable 5.1 and Opus 5 despite being significantly cheaper to use. Zuckerberg said “next up” is Meta’s larger model codenamed “Watermelon,” as well as releasing the weights for Muse Spark. Gemini 3.8 Flash keeps 3.7’s pricing ($0.75 / $3.75) with gains in coding, reasoning, and agentic tasks, slotting in at a 59 on the Intelligence Index. DeepMind’s Koray Kavukcuoglu admitted Gemini sits “a little below the frontier”, and said, “there’s nothing other than being at the frontier that’s important for us.” **Why it matters:** Meta continues to climb with an impressive blend of intelligence and cost (something Google used to own), making Watermelon a potential industry-shaker. For Google, 3.8 is a step in the right direction, but with its Pro model seemingly out of the picture until Gemini 4, there is still some major pressure for a frontier comeback. **The Rundown:** Every enterprise racing to adopt AI eventually hits the same walls: where does the model run, where does the data go, and who can see it. Join experts from CoreWeave and Cosine on Sept. 10 for a practical session for teams deploying AI in regulated environments. **Attend to learn how to:** Define “sovereign” for your org across data residency, deployment, model provenance, and more Turn governance requirements into automated checks with the open-source Responsible AI Toolkit See the full sovereign AI stack come together in a live technical demo **The Rundown:** Each week, Rundown AI educator Nate Grahek breaks down what he's seeing while teaching real teams to use AI. **Nate:** Two weeks ago it was Uber's engineers. Last week it was the AI-pilled people at your small business. Notice what they have in common: they were already technical, or at least really tech-savvy. That's the truth about the ceiling I promised. Here's what changed for me: I stopped nodding along. When a developer mentions a webhook or an environment variable, I don't file it under "not my department" anymore. I turn on voice mode on the walk home and say, "Explain that to me like I run the company but not the codebase." There are no bad questions in that room, and it will re-explain at whatever altitude you ask for. AI raises the floor, but the ceiling stays low if you don't build your tech literacy along with your AI fluency. And I'm talking about a pretty basic level, the way a CEO talks to their CTO. You don't write the code, but you know what a server is, which of your files live in the cloud versus on your laptop, and how to keep them organized enough to find one. If that's already you, keep going. The two skills help each other immensely, and every bit of it compounds. If you closed that door years ago ("I'm not a technical person, I never will be"), here's the great news: there has never been a better time to open it. You now have the most incredible tech support, explainer, and duet partner sitting alongside you. With screen recording or an agent driving your browser, it walks you through things that used to require hours and hours of tech support. So slow down and let AI teach you your own machine. What's one term you've been nodding along to for years? Ask AI to explain it this week, at exactly the level you want. **The Rundown:** In this guide, you will learn how to turn an AI workflow you know into a “proof project” for a job application or interview. You will record a walkthrough, use its transcript to build a five-slide deck, and tailor the finished kit to the role. **Step-by-step:** Pick one AI workflow you know well. Choose something simple but useful, like an AI reporting routine, research process, or content workflow Use a tool like Loom to record a five-minute video of your screen and face. Walk through the AI solution, but emphasize the “human-in-the-loop” aspect Paste the Loom transcript, link, and job description into ChatGPT/Claude. Tell it: “Create a five-slide deck with the video embedded on one slide. Use the other slides to explain my thought process and how the workflow operates” **Pro tip:** Loom has some light editing and AI features, but the goal is to record this in one take without wasting time on edits. **The Rundown:** Apps and agents increasingly need to run long-running background tasks — but serverless is too restrictive, and managing workers and queues by hand is a headache. Render Workflows gives you the best of both in a lightweight SDK. **With Render Workflows, you can:** Define tasks in a few lines of code and chain them into long-running, distributed workflows Launch agents and batch jobs on demand, without spinning up infrastructure Let Render handle queuing, orchestration, and retries automatically [Try it today and get $50 in credits](https://credits-portal-mmdm.onrender.com/claim/rundown-ai-september?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) to run your first workflow. Image source: Images 2.0 / The Rundown **The Rundown:** OpenAI’s upcoming ‘Astra’ model expected this week reportedly [uses](https://www.theinformation.com/articles/secret-technique-behind-openais-astra-model-sparks-security-concerns?rc=a8sc5q&utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) a new technique that provides a boost to its coding and computer-use skills, according to The Information — but also keeps parts of its reasoning out of view for safety monitoring. “Recurrent depth” has the model analyze the same text in repeated loops before answering, squeezing extra intelligence out of the model without making it larger. The loops reportedly improve performance but come with thinking outputs that are pure math, instead of the more readable reasoning in past models. OAI reportedly dialed the loops back so Astra still writes its reasoning out, and its blog [promises](https://openai.com/index/path-to-astra/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) extra monitoring of that reasoning at launch. Chief Scientist Jakub Pachocki posted that he wants to "prevent a race into unmonitorability" from "confused reporting," while calling monitoring "fragile." **Why it matters:** The alarm bells are less about Astra, which OAI says still shows its work, and more about a performance race that tempts the industry to loop harder and monitor less. Pachocki himself admits current monitoring is already fraying for reasons unrelated to Astra, so the main window into what models think may be closing either way. **QUICK HITS** ▸ **John built a custom GPT to explore unified physics equations** Today’s workflow comes from reader **John Gaus:** “I built a custom GPT with the personalities of Einstein, Lorentz, Planck, and Compton. Starting with Einstein’s 1920 Leiden lecture, I developed an ontology for physical space based on his description of the “new ether.” I began with E=mc2, E=hf, and my ontological modeling assumptions. I then pushed ChatGPT to challenge those assumptions and interpretations logically and mathematically. We used several tool calls and reference sites to write and test the math. After many months of working on it in my spare time, the GPT now unifies equations that balance from the atomic scale to the black hole scale, with some remarkable revelations. I summarized much of the work in a paper written by my GPT and am happy to share it so others can expand, improve, and test the work. AI rocks.” *See John’s full workflow here. How do you use AI? Tell us for a chance to be featured.* 🔐 [**Incogni**](https://deal.incogni.io/aff_c?offer_id=6&aff_id=1889&url_id=1&utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) - Erase sensitive data like addresses and phone numbers from the web. Get 55% off with code **RUNDOWN**.\* ⚡️ [Gemini 3.8 Flash](https://www.therundown.ai/tools/gemini-3-8-flash?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) - Google's strongest Flash model yet for coding and agents 🚀 [Muse Spark 1.3](https://www.therundown.ai/tools/muse-spark-1-3?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) - Meta’s new cost-effective, near-frontier model 🎥 [H3 Max Turbo](https://x.com/fal/status/2095210540083884453?s=20&utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) - Fal's video model at twice the speed, half the cost *\*Sponsored Listing* **Get more from Claude** – Your style, voice, and data, supercharged. Unlock Claude’s full potential with custom skills. [Download the guid](https://go.innovatingwithai.com/claude-skills-rundown?utm_source=rundown&utm_campaign=2026-09-03)[e](https://go.innovatingwithai.com/claude-skills-rundown?utm_source=rundown&utm_campaign=2026-09-03)\* **U.S. Commerce Secretary Howard Lutnick** [told](https://www.axios.com/2026/09/02/lutnick-anthropic-trump?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) Axios, “We trust Anthropic,” revealing the company is “back on the right side” of the relationship with the government. **The U.S. government** [filed](https://fingfx.thomsonreuters.com/gfx/legaldocs/jnvwzqxzbpw/OPENAI%20NEW%20YORK%20TIMES%20COPYRIGHT%20LAWSUIT%20govtbrief.pdf?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) a brief in support of OpenAI in the NYT copyright lawsuit, arguing that limiting fair use for AI training would cost the U.S. its global tech lead. **New York City public schools** are [banning](https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) AI till eighth grade, with Mayor Zohran Mamdani rejecting that AI-powered education is "not only inevitable, but necessary." **Spanish AI startup Multiverse Computing** [launched](https://multiversecomputing.com/resources/introducing-quasar-438b-europe-s-leading-ai-model?utm_source=x&utm_medium=organic_social&utm_campaign=quasar_438b_api_0926&utm_content=organic_sep02_article_link) Quasar 438B, its first model, coming in at a 43 on the Artificial Analysis Intelligence Index to top all European rivals. Read our last AI newsletter: [Fable 5.1 kicks off launch week at the frontier](https://www.therundown.ai/articles/fable-5-1-kicks-off-launch-week-at-the-frontier?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) Read our last Tech newsletter: [Dyson puts AI in a $499 toothbrush](https://tech.therundown.ai/p/dyson-puts-ai-in-a-499-toothbrush?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) Read our last Robotics newsletter: [Hugging Face’s robot duck is a hit](https://robotnews.therundown.ai/p/hugging-face-robot-duck-is-already-a-hit?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) Today’s AI tool guide: [Nail job interviews with the “Proof Project” method](https://app.therundown.ai/guides/stand-out-in-an-ai-job-interview-with-the-proof-project-method?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) RSVP to next workshop on Sept. 9: [Build your ad creative strategy with Claude](https://app.therundown.ai/live/build-and-run-your-ad-creative-strategy-with-claude?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-1001) See you soon, *Rowan, Zach, Shubham, Jennifer, and Nate — the humans behind The Rundown*