General AI News (Sep 3): Muse Spark 1.3, Gemini 3.8 & EU Act
Meta went closed-source with Muse Spark 1.3, Google shipped Gemini 3.8 Flash, and the EU sent info requests to 30+ AI companies. The top AI developments today.
Meta went fully closed-source with Muse Spark 1.3, Google shipped its third Flash model in six weeks, and the EU sent information requests to over 30 AI companies. The AI landscape shifted on multiple fronts in the last 24 hours, and the implications for developers, regulators, and enterprises are significant.
Previous Coverage
Missed yesterday's AI news roundup? Read our General AI News from September 2 covering OpenAI's Astra model safety protocols and Apple's leadership transition. Also check our AI Startups from September 2 for AfterQuery's $3.2B valuation milestone.
Meta Ships Muse Spark 1.3, Doubles Down on Closed Source
Meta released Muse Spark 1.3 on September 2, marking the most aggressive step yet away from the company's open-source Llama legacy. The model posted a 75.4% score on DeepSWE 1.1 (an agentic software engineering benchmark), hit 98.5% on long-context retrieval with a 1 million token window, and completed coding tasks with roughly 20% fewer tool calls than its predecessor. Paid API access through the Meta Model API opened September 3.
The strategic pivot matters. Mark Zuckerberg wrote a 6,500-word essay in 2024 titled "Open Source AI is the Path Forward." Muse Spark is the opposite: proprietary weights, no self-hosting, and a monetized API. The Llama family still exists, but Meta has not clarified whether it will continue active development. The company is now shipping on a four-week cadence, directly monetizing its most capable model for the first time.
Bloomberg reported that Muse Spark was built by Meta Superintelligence Labs (MSL), the unit assembled around Alexandr Wang after Meta's $14.3 billion investment in Scale AI. The model still trails Claude Opus 5 on knowledge-work benchmarks like JobBench and OSWorld 2.0, but its coding and long-context performance now leads the field.
Google Releases Gemini 3.8 Flash, Its Third Flash Model in Six Weeks
Google announced Gemini 3.8 Flash on September 2, the company's third Flash-tier release since late July. Ars Technica noted this pace makes it unlikely Google will ship the previously promised Gemini 3.5 Pro. The Flash line targets speed and cost efficiency over peak capability, making it a strong fit for production workloads where latency matters more than benchmark dominance.
A companion model, Gemini 3.8 Flash Cyber, was also released for cybersecurity tasks. It is available through Google Fairwind, a limited-access program for governments and trusted partners. The dual release signals Google's push to serve both commercial and national security customers with specialized variants of the same base architecture.
Artificial Analysis confirmed the model is available on Google AI for Pro and Ultra subscribers, and the developer API is live with standard Gemini pricing tiers.
Anthropic's Claude Fable 5.1 Tops the Intelligence Index
Claude Fable 5.1, released September 1, scored 66 on the Artificial Analysis Intelligence Index at max effort, topping the leaderboard. The bigger story for enterprise users is the pricing: cache reads dropped 75%, from $1.00 to $0.25 per input token. Anthropic also shipped Claude Mythos 5.1 alongside it.
The cache cost reduction is significant for production deployments. Companies running long-running agent loops or RAG pipelines with repeated context windows will see direct cost savings. The Fable line has consistently offered the strongest agentic knowledge-work performance, and the price cut narrows the gap with cheaper alternatives like Muse Spark and GPT-5.6 Sol.
US and EU Clash on AI Regulation at G20
The US and the EU pulled in opposite directions on AI policy at the G20 Innovation Ministerial in Chapel Hill, North Carolina on September 2. Michael Kratsios, Trump's tech adviser, presented the "Carolina Principles," which argue that AI-specific regulations are unnecessary and that governments should regulate by outcomes, not by technology. Meta CEO Mark Zuckerberg and Tesla CEO Elon Musk both spoke in favor of fewer constraints on AI development.
Musk was blunt about Europe: "Things are generally default illegal. It slows it down quite considerably." He also warned of a "significant power shortfall next year" for AI data centers, framing energy infrastructure as the real bottleneck.
On the same day, the European Commission confirmed it had sent information requests to more than 30 AI companies worldwide, a preliminary step that could lead to formal investigations under the EU AI Act. The requests focus on safety and copyright compliance. EU VP Henna Virkkunen wrote on LinkedIn that Brussels is "ready to take all necessary steps" to enforce compliance.
Source: Reuters via Al Jazeera
The divergence creates a compliance headache for companies operating globally. An AI startup serving both markets must navigate one framework that penalizes over-regulation and another that is actively investigating dozens of firms.
Palo Alto Networks Acquires Console for $500M
Palo Alto Networks closed its acquisition of Console on September 2, paying roughly $500 million for the Thrive-backed startup. CEO Nikesh Arora positioned the deal as expanding Palo Alto's reach into markets adjacent to its core security business. The acquisition follows the $25 billion CyberArk deal that closed in February 2026, marking Palo Alto's second major acquisition this year.
Console brings agent-focused infrastructure tooling that Palo Alto plans to integrate into its security platform. The deal reflects a broader industry trend: as AI agents become production systems, the security stack around them is becoming a standalone market.
Enterprises Eye Non-Nvidia Chips Over Next-Gen GPUs
A VentureBeat survey of 170 AI infrastructure respondents found that 39.4% are likely to evaluate non-Nvidia accelerators (AWS Trainium, Google TPU, AMD Instinct, Intel Gaudi, or in-house ASICs) over the next 12 months, compared with just 25.3% planning to evaluate Nvidia's next-generation Blackwell (GB300) GPUs. That 14-point gap is the first time non-Nvidia options have led Nvidia on enterprise evaluation intent.
The shift reflects both cost pressure and availability concerns. Nvidia's data center revenue continues to grow, but enterprises are hedging against a single-vendor dependency. AMD's Instinct MI400 series and Google's TPU v6 are gaining traction in inference workloads where Nvidia's training-focused architecture is less of a differentiator.
Quick Comparison
| Entity | Key Metric | What Changed | Source |
|---|---|---|---|
| Meta Muse Spark 1.3 | 75.4% DeepSWE 1.1 | Closed-source pivot, paid API | Meta AI Research |
| Google Gemini 3.8 Flash | 3rd Flash in 6 weeks | Cyber variant via Fairwind | Google Blog |
| Claude Fable 5.1 | 66 on AI Intelligence Index | 75% cache read price cut | Artificial Analysis |
| Palo Alto / Console | $500M acquisition | Agent-focused security tooling | TechCrunch |
| EU AI Act enforcement | 30+ info requests | Safety and copyright focus | Al Jazeera |
Frequently Asked Questions
What is the top AI news today?
Meta released Muse Spark 1.3, its most capable closed-source model, scoring 75.4% on DeepSWE 1.1 and opening paid API access on September 3. Google shipped Gemini 3.8 Flash, its third Flash model in six weeks, while the EU sent information requests to over 30 AI companies for AI Act compliance checks. Anthropic's Claude Fable 5.1 topped the Intelligence Index with a 75% cache cost reduction.
What did Meta announce with Muse Spark 1.3?
Meta released Muse Spark 1.3 on September 2, a closed-source multimodal reasoning model with a 1 million token context window and 75.4% on the DeepSWE 1.1 coding benchmark. The model trails Claude Opus 5 on knowledge-work tasks but leads on coding and long-context retrieval. Paid API access through the Meta Model API opened September 3, making this Meta's first directly monetized frontier model.
How did the EU and US differ on AI regulation?
At the G20 Innovation Ministerial on September 2, the US pushed the "Carolina Principles" against AI-specific regulation while the EU simultaneously sent information requests to 30+ AI companies for AI Act compliance. Elon Musk criticized European regulation as making things "default illegal," while EU VP Henna Virkkunen warned Brussels is "ready to take all necessary steps" to enforce its framework.
Related reading
If this story matters to you, these earlier editions cover the same ground in more detail:
- General AI News (Sep 21): UN Agent Warning, Gemini Hack & Huawei
- General AI News (Oct 1): Gemini 4 Argon, Broadcom's $42B & FTC
- Z.ai Opens GLM-5.3-Flash, Google Ships Gemini 3.5 Transcribe, Perplexity Goes Local: General AI News for August 28, 2026
- Viral AI News (Oct 3): EBAI, ds4 & Muse Gadgets
Building an AI agent, automation, or backend workflow? AsaanKaroobar builds custom AI solutions, apps, and digital experiences. Get in touch →
