BIG TECH

Meta gets back in the game with Muse Spark model

Meta finally stopped talking about releasing new AI models and actually released one. 

On Wednesday, the company debuted Muse Spark, the first model from the company’s superintelligence lab. Meta claims this model is purpose-built to “prioritize people,” focused specifically on powering Meta AI. 

The company will offer the model in private preview to “select partners,” and hopes to open-source future versions of the model. Muse Spark will roll out to WhatsApp, Instagram, Facebook, Messenger, and AI glasses in the coming weeks. 

Here are some of the specs: 

• The model accepts multimodal inputs, including voice, text and images, but only outputs text. The model also includes a “shopping mode” that allows users to surface products based on prompts. 

• The company said the model is “small and fast by design,” but capable of reasoning across complex queries in science, math and health and built to support “complex reasoning and multimodal tasks.” 

• According to Artificial Analysis, Muse Spark stacks up to the likes of Gemini 3.1 Pro, GPT-5.4 and Claude Opus 4.6 in common benchmarks, sitting in fourth place on the Artificial Analysis Intelligence Index, ahead of models from xAI, Z.ai and DeepSeek.  

Meta said that Muse Spark is its most powerful model yet, but remains an “early data point” in its scaling trajectory. It’s the first in its new series of large language models, according to the press release, with “larger models in development.” 

“We are on our way to personal superintelligence: an assistant that can help anyone, anywhere with the things that matter most to them,” Meta said in its announcement. 

The model, initially codenamed Avocado, is the first released under Alexandr Wang, the company’s chief AI officer, after it acqui-hired him from Scale AI. It’s the product of nine months of work after Meta rebuilt its AI stack “from scratch,” Wang said in a post on X. It marks a massive improvement from its Llama model family, and a pivot away from its purely open model roots.

With Muse Spark, Meta is finally showing its teeth. The company has spent billions amassing talent over the past year, poaching from the likes of OpenAI and Google DeepMind, offering bonuses of up to $100 million dollars to lure some in. Muse Spark seems to be the first evidence that its spending spree has gone to good use. Still, while Muse Spark is competitive with leading rivals, the company has not claimed that the model is state of the art. With next-generation frontier models on the horizon, such as Claude Mythos and OpenAI’s “Spud,” Meta may just be trying to get in the game again after spending months on the bench.

TOGETHER WITH PIGMENT

AI maturity is driving a 12-point growth gap

Most teams react to uncertainty after it shows up in results. Pigment’s Uncertainty Index reveals how AI is separating leaders from laggards. 

Companies with mature AI capabilities grew 18.1% last year, compared to just 6.2% for early-stage peers. That gap shows up in how teams operate too: AI leaders reforecast nearly twice as often, turning uncertainty into faster, data-driven decisions. Yet many executives still overestimate their AI readiness. For AI-focused teams, the takeaway is clear: better outcomes don’t come from models alone, but from how deeply AI is embedded in decision-making. Explore full insights.

Download here

PRODUCTS

S26 Ultra previews the agentic phone era

The biggest barrier to AI adoption isn't the technology itself. It's having to use yet another tool, and smartphones with built-in AI offer a solution. The question is whether they can bake in features that are genuinely helpful and worth using every day. I spent the past month testing the Galaxy S26 Ultra, focusing just on that.

But before I run through the list of all of the AI features, it is only right that I dive right into perhaps the one people have been most excited about: the task automation feature powered by Google Gemini. It's essentially an AI agent for your smartphone. The verdict? It does indeed work, exactly as intended.

For instance, if I tell it to order me an Uber to a specific address and for a specific class of car, such as “Order me an Uber Comfort to 123 Boyfriend Lane,” and it is able to click through all the options, so all I have to do is approve it. It doesn't exactly save you time, but the real magic is that while Gemini's busy clicking buttons, you're free to do other things — like finding your keys.

While Task Automations is the flashiest feature, others are a testament to an important, often-overlooked bucket of AI features: the ones that are helpful but you forget are even there. You can find a quick rundown below: 

• Document scanning: Allows users to scan images by simply pointing their phones at them, while AI ignores creases, removes distortions, and completes the photo if it has bent corners. 

• Now Nudge: This feature provides real-time suggestions across any messaging app, integrated into the keyboard. In my experience, the feature kicked in by day two and felt like enhanced autocomplete, making it subtly quicker to send messages. 

• Audio Eraser: The audio eraser function is super neat because it lets you reduce background noise in real time on apps like YouTube, Netflix, Instagram, and TikTok. In my experience, it was easy to activate and worked exactly as promised.

• Photo Assist: While many Android users already rely on Google Photos, which has had native AI photo editing features for a while, it is convenient to access more advanced AI photo editing tools natively on your device, which you can use to touch up the photos even before you back them up to Google Photos.

For the full review, including a deeper dive into the AI features, as well as the hardware (including the super handy new Privacy Display), check out the full story on our website.

As seen with nearly every phone launch since 2023, AI has been at the forefront of device releases. Since then, we have seen both ends of the spectrum: the Google Pixel 10 earning wide recognition as the first truly competitive AI smartphone, and Apple drawing significant criticism for announcing a suite of Apple Intelligence features that have yet to launch in the iPhone. The Samsung Galaxy S26 lineup followed the Pixel 10's lead, offering subtle, baked-in AI capabilities that hold up in everyday use, while also introducing a flashier agentic feature that works as intended, even if it's a modest start. From an AI perspective, this is a well-thought-out approach.

TOGETHER WITH COMPOSIO

Take your AI agent everywhere you work

The smartest model in the world is useless if it can't actually do anything.

Composio connects your agents to 1000+ apps - Gmail, GitHub, Salesforce, and more - so they stop answering questions and start executing real workflows across every platform you use.

Use it right inside Claude, Cursor, or Notion AI - or build with the SDK.

Join hundreds of thousands of users in leveling up your agentic workflows.

Get started for free.

SAFETY

AI agents increase the cost of being wrong

What happens when a startup's mission perfectly aligns with the biggest trend in tech? That's exactly where Guild AI finds itself in 2026.

James Everingham, CEO of Guild AI, joins The Deep View Conversations to talk about building a safety layer for AI agents. The product launched in fall 2025 and found itself at the center of the most important movement in enterprise AI just months later.

In this conversation, James breaks down how Guild's platform deploys dozens of workflow-specific agents across different parts of a business, while giving developers the tools to iterate, spin up custom agents, and operate in a safe environment that tracks everything agents do and protects companies from unpredictable outcomes.

Topics covered:

• The enduring power of bottom-up innovation

• How Guild AI's agent supervision platform works

• Why safety infrastructure is the new competitive moat

• Lessons from James' earlier career at Netscape and Meta

• Open-source vs. proprietary models: how it plays out over the next few years

• A standout leadership tip for sparking innovation on your team 

• His best advice for getting maximum impact from today's AI tools.

If you want to understand where AI agents are headed and what it takes to build them responsibly, this conversation is a powerful place to start.

Subscribe for weekly conversations with the leaders shaping the future of AI.

LINKS

• Alibaba and China Telecom launch data center in China for AI training

• X rolls out automatic post translation worldwide, powered by Grok

• Perplexity’s ARR rose to $450+ million, according Financial Times report

• Intel and SambaNova announced a new blueprint for agentic AI 

• Visa unveiled Intelligent Commerce Connect for agentic commerce

• Bain’s Data Center unit drops firm over suspected Nvidia smuggling

• OpenAI releases Child Safety Blueprint to prevent sexual exploitation

• Z.ai: The company launched GLM 5.1, outperforming Opus 4.6 and GPT-5.4 

• Claude: New Claude Managed Agents makes it easier to build and deploy agents 

• Stability.ai: New Brand Studio platform aims to be a creative production platform 

• Runway: The Runway Characters feature now supports camera and screen sharing

• Google: Senior Research Scientist, AMO Modeling, Quantum AI

• LG: Embodied AI Researcher

• Physical Intelligence: Research Scientist

• Microsoft: Senior Applied Scientist

GAMES

POLL RESULTS

Are you worried about AI fueling job loss?

Yes (48%)
Somewhat (35%)
No (12 %)
Other (5%)