AI NewsMeta

Meta opened Muse to third-party developers, letting outside services build connectors so users can reach them just by asking their agent, ten days after Muse's public launch.
Mark Zuckerberg announced on September 18 that developers can now build connectors for Muse, letting users access outside services just by asking their agent. Submissions go through Meta's review for functional, security, and legal requirements before appearing in Muse's directory.
Meta also added Stripe Link for payments, pointing toward connectors that complete transactions, not just read and summarize. It's a fast pivot: Muse launched just ten days earlier, and Meta is already racing to make it the default way people reach other apps.
Key Capabilities:

• Developer submission portal: Businesses submit connectors through a three-step review process.
• Payment-ready: Stripe Link integration points toward transactions, not just read-only tasks.
• Fast rollout: Opened to developers just ten days after Muse's public consumer launch.

Meta released a dedicated Mac app for its personal AI agent Muse, letting it organize downloads, find lost files, and summarize notes across your computer's apps with approval required for sensitive actions.
Meta released a dedicated Mac app for Muse on September 17, letting the agent organize your downloads folder, find a lost file, and summarize messages and notes across your computer. It runs in a secure virtual machine and always asks approval before sensitive actions.
It's currently US-only, with free and paid tiers, building on Muse's iOS, Android, and web versions from earlier this month. Early users give it five-star reviews, calling it the best personal assistant they've tried for everyday tasks like booking or chores.
Key Capabilities:

• Cross-app control: Organizes downloads, finds lost files, and summarizes notes and messages.
• Approval-gated: Always asks before taking any sensitive action on your files.
• More coming: Currently US-only, with additional Mac features planned soon.

Meta launched Meta One, a subscription spanning Facebook, Instagram, WhatsApp, and Meta AI, with plans from $2.99 to $19.99 a month unlocking more AI image and video generation.
Meta launched Meta One on September 15, bundling paid perks across Facebook, Instagram, WhatsApp, and Meta AI. Plans start at $2.99 a month, running up to $19.99, and unlock more AI image and video generation plus 50+ extra features.
The core apps and Meta AI stay free for everyone, so this is about monetizing power users, not gatekeeping access. It's part of a broader push: Meta's shares jumped over 10% in the past week amid the rollout.
Key Capabilities:

• One subscription, four apps: Bundles perks across Facebook, Instagram, WhatsApp, and Meta AI.
• AI generation upgrade: Higher-tier plans unlock more Muse image and video generation.
• Free core stays free: The base experience on every app and Meta AI remains unchanged.

Meta's Muse Spark 1.3, released September 2, reportedly jumped on the independent Agent Arena leaderboard this week and scores 61 on Artificial Analysis's Intelligence Index, tying top rival models at a lower cost.
Meta released Muse Spark 1.3 on September 2. It reportedly jumped on the independent Agent Arena leaderboard this week and scores 61 on Artificial Analysis's Intelligence Index, tying GPT-5.6 Sol and Grok 4.6.
That ties top models at a lower cost, and Meta says it asks more clarifying questions while cutting unneeded tool calls by 20%. On Meta's own chart it still loses most head-to-head agent tests to Claude Opus 5.
Key Capabilities:

• Efficiency gains: Uses 25% fewer tokens and cuts unnecessary tool calls by 20%, per Meta.
• Priced to compete: Available via Muse Code subscriptions starting at $5 a month.
• Mixed picture: Wins coding and long-context benchmarks but loses most agent tasks to Claude Opus 5.

 It jumped 5 positions to take the #1 spot on Website Arena, a month after Muse Spark 1.2's release.
Design Arena reported Muse Spark 1.3 (xhigh) took first overall on Website Arena with an Elo of 1362, a jump of 5 positions from Muse Spark 1.2. This establishes a new Pareto frontier for speed and price in website generation.
It's only been a month since Muse Spark 1.2 topped this same category, showing Meta's rapid iteration pace. Design Arena notes GPT-6 Astra's results are still pending in this specific category.
Key Capabilities:

• Top Ranked: Scored an Elo of 1362, the highest on Website Arena among all tested models.
• Rapid Improvement: Jumped 5 positions above its own predecessor, Muse Spark 1.2, in just one month.
• New Frontier: Sets a fresh benchmark for the balance of speed and price in website generation.

 Muse Spark 1.3 delivers strong agentic coding performance at a fraction of the cost of pricier rivals.
Meta's Muse Spark 1.3 became the first Meta model integrated into Cursor, the AI-powered code editor, just days after its September 2 release. It supports up to 1 million token context windows and tool use across files and the web, built for challenging, long-horizon coding tasks.
It quickly gained traction, processing 23 trillion tokens on OpenCode and topping charts like CursorBench at 67.9% for $1.31 per task, matching pricier rivals like GPT-5.6 Sol at a fraction of the cost. One developer reportedly finished a full project in two hours for about $1.
Key Capabilities:

• Strong Price-Performance: Scores 67.9% on CursorBench for just $1.31 per task on average.
• Massive Context: Supports up to 1 million token context windows for large, complex codebases.
• Rapid Adoption: Processed 23 trillion tokens on OpenCode within days of Cursor integration.

Muse runs 24/7 in a secure virtual machine, pursuing your long-term goals with your approval.
Meta launched Muse, a personal AI agent that connects to your apps to send emails, fill forms, and make purchases with your approval, working around the clock to get things done across every part of life. It runs in a secure, isolated virtual machine that keeps your data and credentials separate from the model itself.
A built-in review system checks actions before they go live, and Meta stresses user control over permissions at all times. It's available now in the US on iOS, Android, and web, with Meta's AI glasses support coming soon, offering both free and paid tiers.
Key Capabilities:

• Isolated Execution: Runs in a secure virtual machine, keeping credentials separate from the AI itself.
• Action Review: Checks consequential actions with the user before executing them online.
• Cross-Platform: Available now on iOS, Android, and web, with AI glasses support coming soon.

 AIRA₃ placed 8th out of about 4,000 teams, earning gold in an NVIDIA-run Kaggle contest.
Meta's autonomous AI research system AIRA₃ entered a live Kaggle competition run by NVIDIA to fine-tune a 30 billion parameter Nemotron model for better reasoning. All competitors had access to the same information and were graded externally on a private test set.
AIRA₃ placed 8th out of roughly 4,000 teams, winning gold and outperforming human competitors using the same frontier tools. It runs many long-running agents that coordinate asynchronously through a shared forum and filesystem rather than a central controller. The same system also cut latency on GPU kernels by 27% and translated ancient Akkadian tablets.
Key Capabilities:

• Beat Human Experts: Placed 8th of nearly 4,000 human teams in an externally graded live competition.
• Multi-Agent Design: Coordinates many long-running agents through a shared forum instead of one controller.
• Generalizes Well: Also cut GPU kernel latency by 27% and handled ancient language translation tasks.

 Muse Spark 1.3 Max shows significantly stronger coding and agentic performance than earlier Spark versions.
Meta publicly released Muse Spark 1.3 Max, showing significantly stronger coding and agentic performance than Muse Spark 1.3 High or XHigh. On the updated Artificial Analysis Intelligence Index v4.2, it still performs strongly, with the efficient frontier now dominated by Muse, Claude, and GPT models.
Developers testing it against pricier rivals report being surprised by the quality, with one noting it could pass as Opus 4.8 in a blind test. It's positioned as a strong low-cost pick for everyday coding work.
Key Capabilities:

• Strong Efficiency: Sits on the efficient frontier alongside top Claude and GPT models on cost versus performance.
• Surprising Quality: Developers report it rivals pricier models like Opus 4.8 in blind comparisons.
• Multiple Tiers: Builds on Muse Spark 1.3 High and XHigh with a stronger Max variant.

Muse Spark 1.3 sustains longer work sessions while using 20% fewer tool calls than its predecessor.
Meta released Muse Spark 1.3 with improved performance on agentic and coding tasks, focused on real world usability. It sustains longer horizon work across multiple workflows in a single thread and more actively collaborates with users by asking clarifying questions when stuck.
It's better calibrated on its own limits instead of hallucinating outcomes, using about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2. Zuckerberg called it the biggest jump Meta has made so far on coding and agentic work.
Key Capabilities:

• Smarter Collaboration: Asks clarifying questions and confirms before taking consequential actions.
• More Efficient: Uses roughly 20% fewer tool calls and 25% fewer tokens than the previous version.
• Better Self-Awareness: Calibrated to recognize its own limits rather than hallucinating confident answers.

Atlas generates image and video frames with pixel-perfect camera control, then reconstructs them in 3D.
World Labs introduced Atlas, described as the world's first multimodal world model that generates image and video frames with precise camera control and reconstructs them into full 3D scenes. Users can move the camera and simulate space and time within generated environments.
This pushes AI generation beyond flat images or clips into interactive, navigable 3D spaces. It signals a new direction for AI models built around spatial understanding rather than just pixels.
Key Capabilities:

• Precise Camera Control: Generates frames with pixel-perfect camera movement, not just static images.
• 3D Reconstruction: Rebuilds generated scenes into full 3D environments you can move through.
• World Simulation: Models both space and time, not just single images or short clips.

Muse Code now supports subagent workflows and sessions that message each other, starting at $5 a month.
Meta moved Muse Code out of beta, adding sessions that message each other and workflows that orchestrate teams of subagents. An SDK is now in developer preview for building custom agents on top of it.
Monthly plans start at $5, scaling up to $50 for power users with expanded limits. Developers can install it with a single terminal command on macOS and Linux.
Key Capabilities:

• Multi-Agent Workflows: Splits complex engineering tasks across coordinated subagent teams working together.
• Session Messaging: Different coding sessions can now message each other to share work seamlessly.
• SDK Preview: Developers can build their own custom agents on top of Muse Code's foundation.