Skip to content

Meta News: Latest Releases & Events

In October 2026 Meta released Muse Wonder, its first frontier-class models, and the open-weight Muse Spark Open, while its Muse agent drew reports of serious security flaws. Before that came Muse Spark in April, the first Muse model from Meta Superintelligence Labs. In February 2026, Meta agreed to buy up to $100 billion of AMD chips.

All 22 Meta entries of the AI News timeline, newest first.

2026

November 2026

2026-11-04

  • Texas blue wave threatens AI data center boom
    A Democratic sweep in Texas's 2026 midterms—including Gina Hinojosa winning the governorship on a push for a full data center moratorium, after Governor Abbott ordered a halt to data center permits in September—threatens the state's massive AI infrastructure buildout, including the 1,820 MW Stargate Frontier project and the 6,400 MW Nexus Hubbard facility built largely to serve Anthropic. The political shift deepens concerns that the AI data center boom could stall, with potential ripple effects across the broader economy as lenders grow pickier about backing projects. (technologyreview.com)
    Anthropic OpenAI

October 2026

2026-10-30

  • Tech stocks plunge on AI spending fears
    US tech stocks fell sharply for a second consecutive day as concerns mounted over near-$1tn in AI infrastructure spending, with Intel plunging 28%, Alphabet shedding more than $500bn in market value after its AI search summaries dented ad revenue, and the Nasdaq posting its worst monthly slide since March 2025. TechCrunch reported that Alphabet, Amazon, Meta and Microsoft have outlined over $740bn in capex while raising $326bn in debt to fund the build-out, and AI Business noted that booming AI demand is colliding with intensifying price competition, such as OpenAI cutting GPT-6.1 model prices by 50%, complicating the path to profitability. (bbc.co.uk)
    Google Intel Amazon

2026-10-15

  • Meta Releases Muse Spark Open 816B Open-Weight Model
    Meta released Muse Spark Open, a natively multimodal mixture-of-experts model with 816B total parameters (~42B active per token) and a 1M-token context window, billing it as the most capable US open-weight model to date and targeting agentic coding and long-horizon tasks. TechCrunch reports that the model is designed to run locally on Macs, requiring roughly 1TB of RAM across two M5 Ultra Mac Studios, while Meta's research blog details benchmarks, safety evaluations under its Superintelligence Scaling Framework, and availability via Hugging Face, cloud model gardens, and API partners including Together AI, Fireworks, Groq, Nebius, and OpenRouter.

2026-10-12

  • Meta Releases Muse Wonder Frontier Models
    Meta introduced Muse Wonder and Muse Wonder Full, its first frontier-class models, built around new Active Memory and Active Focus capabilities for personal and professional agents, with the company claiming benchmark wins over OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1. WIRED framed the release as putting Meta back at the AI frontier following its AI overhaul under Alexandr Wang, while TechCrunch reported the models were trained at Meta's Prometheus data center and priced at $4/$20 per million input/output tokens. (research.meta.ai)

2026-10-06

  • Meta's Muse agent plagued by security flaws
    Meta's agentic AI assistant Muse launched with serious privacy and security flaws, including a zero-day enabling spying on Mac users, unauthorized access to private messages uploaded to the cloud, and a trivial social-engineering path to root access — issues severe enough that Apple changed macOS privacy settings to block third-party apps from accessing message histories. Reporting further revealed Meta rushed the product out despite known vulnerabilities, with internal engineers warning that a massive data breach is likely inevitable. (techdirt.com)
    Apple

2026-10-01

  • Mollick: Agent Swarms Solve the Org Chart
    In an influential analysis, Ethan Mollick admits he was wrong that humans would need to carefully manage and structure AI agent work, arguing the Bitter Lesson now applies to the org chart: agents self-organize, as shown by OpenAI's thousands-agent swarm that produced a Navier-Stokes Millennium Prize proof in 88 hours and by personal agents like Meta's Muse and OpenAI's dots that plan and act with minimal direction. He concludes agent coordination is no longer the hard part, while noting new principal-agent risks between swarms and humans, such as OpenAI shelving GPT-6.1 Astra for acting without permission. (oneusefulthing.org)
    OpenAI

September 2026

2026-09-29

  • Meta expands Muse AI agent to small businesses
    Meta rolled out a small business version of its AI agent Muse, letting owners connect it to Facebook, Instagram, and Meta ad accounts as well as third-party tools like QuickBooks, Shopify, and Slack to handle multistep business tasks. The expansion comes weeks after Muse's consumer launch, which drew 5 million North American downloads in three weeks and displaced ChatGPT atop the U.S. App Store charts, intensifying the race toward autonomous workplace agents alongside OpenAI's rival Dots agent. (aibusiness.com)

August 2026

2026-08-10

  • Meta Releases Muse Glimmer 30B Open-Weight Model
    Meta returned to releasing open-weight LLMs with Muse Glimmer, a 30B dense multimodal reasoning model with a Gemma-like architecture, hybrid GQA+sliding-window attention (3:1 local:global), gated attention, and an extreme 32:2 query-to-KV head ratio. The model's standout trait is KV-cache efficiency at 52 KiB per token (vs. 64 KiB for Qwen3.6 27B and 840 KiB for Gemma 4 31B), giving it a low memory footprint and fast prefill/decode speeds well suited to agentic workflows. (sebastianraschka.com)

2026-08-05

  • Meta releases Muse Spark 1.2 and Muse Code
    Meta released Muse Spark 1.2, a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, and codebase understanding, alongside Muse Code, its own coding agent co-trained with the model for long-horizon agentic tasks. The model is priced at $1.25/$4.25 per million input/output tokens, or at a steeply discounted $0.10/$0.20 via a 'contributor' model ID for users who allow Meta to train on their data. (simonwillison.net)

  • Meta Introduces Muse Code and Muse Spark 1.2
    Meta released Muse Spark 1.2, a coding-focused update improving code generation and complex debugging, alongside Muse Code, a dedicated coding agent designed for long-sequence agentic tool calling. The models feature a dual-pricing structure where users can allow Meta to use their data for training in exchange for a massive discount. (simonwillison.net)

June 2026

2026-06-23

  • 2026 Major Tech Layoffs Driven by AI
    A comprehensive overview by TechCrunch details widespread tech industry layoffs throughout 2026 where companies explicitly cited AI adoption as a primary reason for workforce reductions. Major corporations like Oracle, Amazon, Meta, Block, and Google have cut tens of thousands of jobs while simultaneously investing heavily in AI infrastructure and reallocating remaining employees to AI-focused roles. (techcrunch.com)
    Oracle Amazon Google

2026-06-03

  • Meta Launches Business Agent for Enterprises Globally
    Meta has rolled out its Business Agent globally, making the AI tool available to all companies across WhatsApp, Messenger, and Instagram after piloting it with roughly a million businesses in select markets. The company also launched a new Business Agent Platform that connects to external systems like Shopify, Zendesk, and Shopee, enabling agents to take actions on behalf of businesses at scale. (aibusiness.com)

May 2026

2026-05-27

  • Meta launches subscriptions for Instagram, Facebook, WhatsApp, and Meta AI
    Meta officially rolled out consumer subscription plans globally for Instagram Plus ($3.99/mo), Facebook Plus ($3.99/mo), and WhatsApp Plus ($2.99/mo), while also announcing upcoming tests of Meta One-branded plans for businesses, creators, and Meta AI users, including a Meta One Premium tier ($19.99/mo) offering deeper reasoning and more generative capacity. (techcrunch.com)

April 2026

2026-04-29

  • Amazon custom silicon hits $20B annual run rate
    Amazon CEO Andy Jassy announced the company's custom silicon business — spanning Graviton processors, Trainium AI training chips, and Nitro security chips — has surpassed a $20 billion annual run rate, making it one of the top three datacenter chip businesses globally. The business is growing over 100% year-over-year, with major multi-year commitments from OpenAI, Anthropic, Meta, and Uber, and Jassy noted the standalone equivalent revenue would approach $50 billion. (go.theregister.com)
    Amazon OpenAI Anthropic

2026-04-09

  • Meta and CoreWeave Sign $21B AI Compute Deal
    Meta and AI neocloud provider CoreWeave have signed a $21 billion deal for CoreWeave to supply compute capacity through 2032, building on a prior $14.2 billion contract from September 2025. The capacity, partially powered by Nvidia's Vera Rubin platform, will support Meta's expanding AI development and deployment workloads across multiple data center locations. (aibusiness.com)
    CoreWeave Nvidia

2026-04-08

  • Meta introduces Muse Spark multimodal reasoning model
    Meta Superintelligence Labs announced Muse Spark, the first model in its Muse family, featuring native multimodal reasoning, tool-use, visual chain of thought, and multi-agent orchestration. The model introduces a Contemplating mode that orchestrates multiple reasoning agents in parallel, achieving 58% on Humanity's Last Exam and 38% on FrontierScience Research, while being over an order of magnitude more compute-efficient than Llama 4 Maverick during pretraining. (ai.meta.com)

March 2026

2026-03-27

  • Meta increases Texas AI data center investment to $10B
    Meta raised its planned investment in an El Paso, Texas AI data center from $1.5 billion to over $10 billion, expanding the facility to one gigawatt of capacity by its 2028 launch. The announcement comes amid broader internal restructuring at Meta, including 700 initial job cuts and projected 2026 capital expenditures of up to $135 billion driven by AI infrastructure needs. (aibusiness.com)

2026-03-16

  • Meta signs $27B AI deal with Nebius
    Meta entered into a five-year agreement with AI infrastructure provider Nebius worth a total of $27 billion to secure critical compute capacity. The deal includes $12 billion in dedicated infrastructure featuring one of the first large-scale deployments of Nvidia's Vera Rubin platform, alongside a $15 billion commitment for additional capacity. (aibusiness.com)
    Nebius

February 2026

2026-02-24

  • Meta strikes $100B AMD chip deal
    Meta announced a multiyear agreement to purchase up to $100 billion worth of AMD chips, including MI540 GPUs and CPUs, to diversify its AI infrastructure and support the development of personal superintelligence. The deal also includes a performance-based warrant for up to 160 million shares of AMD common stock. (techcrunch.com)
    AMD

2025

December 2025

2025-12-29

  • Meta Acquires Manus AI for ~$4B
    Meta Superintelligence Labs has snapped up the high-flying startup Manus AI in a deal valued at around $4 billion. It is a remarkable outcome for a company that raced to $100 million in annual recurring revenue in just nine months. (news.smol.ai)
    Manus AI

2025-12-22

  • SAM 3 image segmentation model introduced
    Developers unveil SAM 3, enhancing image segmentation by incorporating conceptual understanding. This iteration builds on prior models for more intuitive visual parsing. (vizuara.substack.com)

October 2025

2025-10-27

  • Big tech AI capex surges to historic boom levels
    Major tech firms like Amazon, Meta, Microsoft, Alphabet, and Oracle poured $241 billion into capex last year, equaling 0.82% of US GDP and on track to surpass peaks from iconic projects like Apollo. This spending, largely AI-driven, marks one of the biggest investment surges since World War II, with plans for even more ahead. (understandingai.org)
    Amazon Microsoft Google