This week in AI? 

Big Tech cat fights, major model releases and juicy leaks galore. 

Anthropic and OpenAI went full catfight with back-to-back Super Bowl ads AND back-to-back model releases within an hour of each other.

SpaceX and xAI became one mega-company.

Meta? After almost a year of total silence, leaks suggest they're reportedly about to drop everything at once.

What'd you miss?

(If you're reading us daily, you already know. Gold star for you.)

But if life got busy and you need the quick version, here's the AI news that matters this week.

This story didn't get a ton of headlines. But it could be one of the most impactful launches from OpenAI in a while.

OpenAI introduced Frontier this past week, a new platform designed for enterprise customers to build, deploy, and manage AI agents from both OpenAI and third party companies.

The goal? Tackle what OpenAI calls "agent sprawl" by unifying fragmented tools, disconnected workflows, and siloed data.

Each AI agent on the platform gets a unique identity along with permissions and guardrails, helping enterprises maintain security and compliance.

Frontier supports agents across local, cloud, and OpenAI hosted environments. Early adopters include Intuit, HP, Oracle, and Uber.

OpenAI is also partnering with Abridge, Clay, Ambience, Decagon, Harvey, and Sierra to refine the platform based on real customer needs.

What it means: Right now, this is limited to the largest enterprise companies in the world.

But whatever OpenAI cooks up in Frontier will eventually get rolled out across ChatGPT and their other offerings.

What happens with the biggest companies will eventually impact hundreds of millions of OpenAI users.

How do you spend millions on a Super Bowl ad just to roast your biggest competitor?

Ask Anthropic.

Anthropic's first ever Super Bowl commercial ignited a public debate about advertising in AI chatbots and drew a direct response from OpenAI CEO Sam Altman.

The ad promised to keep Claude ad-free, contrasting with OpenAI's recent announcement that ChatGPT will begin showing ads on its free tier and more affordable $8 a month ChatGPT Go plan.

The ads depicted scenarios like gym coaching and therapy sessions where the AI suddenly starts pushing unrelated ads mid-conversation.

Funny? Yuuuuup.

Accurate to how OpenAI's ads will actually work? Not really.

Which is why Altman clapped back on X, calling the ads dishonest and saying OpenAI would never run ads as depicted.

Anthropic declined to comment but explained in a blog post that its revenue comes from enterprise contracts and subscriptions.

What it means: Anthropic put their future on the line with this one.

Let's be honest. It's hard to see how they can hold to a no-ads promise over the long term as inference costs drop and open source competitors close the gap on their coding and API advantages.

Shares of major software and data services companies got absolutely wrecked this week.

Anthropic launched new industry specific AI plugins for its Claude Cowork platform, offering specialized tools for sales, finance, data, marketing, and legal sectors.

These plugins let users automate tasks like file management, document drafting, and folder organization.

The finance and legal plugins in particular shook the market. The main software industry ETF dropped almost 6% on Tuesday, its single worst day decline since April.

Thomson Reuters stock suffered its biggest one day loss ever, falling more than 15%. LegalZoom tumbled almost 20%.

Investors are worried that companies may now rely on AI powered plugins instead of paying for multiple expensive external software tools, directly threatening the software as a service business model.

Analysts are cautioning that the sell-off is mainly driven by uncertainty and suggesting the market may recover as long term effects become more clear.

What it means: Professional services, especially high priced ones in legal and finance, are going to go through a complete makeover in 2026.

Expect a new class of AI native companies doing finance, legal, and consulting work. The big players either change their business model or start getting gobbled up.

The shock waves are just starting.

According to Bloomberg, Elon Musk has combined SpaceX and xAI, forming one of the world's largest private companies at a privately valued $1.2 trillion.

This new joint effort comes as SpaceX prepares for a highly anticipated IPO later this year, competing with expected public offerings from OpenAI and Anthropic.

The merger gives SpaceX access to xAI's AI capabilities while xAI benefits from SpaceX's resources and a significant cash infusion.

Musk announced plans to launch a constellation of a million satellites for orbital AI data centers, aiming to make space the lowest cost way to generate AI compute within two to three years.

Industry experts say the idea is technically possible but not really feasible with today's technology. It will likely take many more years to become practical.

The combined company now includes AI, rockets, space based Internet, and the social media platform X, which xAI acquired last March.

What it means: Most people saw this kind of bundling coming from Elon Musk.

Some analysts suggest the merger is a strategic narrative move to boost SpaceX's valuation ahead of its IPO, framing it as a leader in both space and AI.

The real question is whether investors buy into this big vision.

After a $14 billion acquihire of Scale AI and a complete shakeup of their internal AI teams, Meta has been completely silent on the AI release front for over ten months.

Welp, that's about to change.

Testing Catalog reports that Meta AI is preparing a sweeping platform upgrade with new agents, deeper integrations, and a new large language model.

The acquisition of Manus AI, completed last month, is leading to a dedicated Manus agent and a browser agent reportedly called Sierra.

There's reportedly an OpenClaw integration that would let users connect any AI model with their own API key. A new Avocado AI video model is being readied in both standard and thinking modes.

Meta is also adding a widget prompting users to link Gmail, Google Calendar, and Outlook, signaling an MCP connector could go live soon.

These updates are expected to roll out possibly as early as this quarter.

What it means: Internal code shows Meta is benchmarking Gemini, ChatGPT, and Claude for possible multi-model routing or performance comparison.

After the $14 billion Scale AI deal and the Manus acquisition, Meta is clearly loading up for a big push. The timing could determine their competitive position as rivals launch new models left and right.

Perplexity announced this past week that it's launching Model Council, a feature that runs a single query across multiple frontier AI models at once and synthesizes the results into one response.

The Model Council automatically compares outputs from models like Claude Opus 4.6, GPT 5.2, and Gemini 3.0.

It highlights where the models agree or disagree, removing the need to manually cross-check performance.

Perplexity says their internal data shows AI model performance varies widely by task. Some models excel at coding while others perform better at research or creative work.

The feature is positioned for high stakes use cases like investment research and fact verification where a single model's blind spots could lead to costly mistakes.

Model Council is only available to Perplexity Max subscribers on the web, with mobile support coming soon.

Perplexity also launched an upgraded deep research tool and announced a new internal benchmark called Draco.

What it means: This "mixture of models" concept was predicted, and now a major player is shipping it.

It could fundamentally change how people verify AI generated information.

The catch? You've gotta be on Perplexity's highest paid tier to use it right now.

Nah, Anthropic wasn't done making noise this week.

Anthropic released Opus 4.6, its most advanced AI model to date, upgrading from Opus 4.5 which debuted back in November.

The standout feature is agent teams, which lets multiple AI agents split and coordinate complex tasks working in parallel to finish jobs faster. Agent teams is available as a research preview for API users and subscribers.

Opus 4.6 now supports a context window of a million tokens on the API side, allowing it to process much larger volumes of information.

But here's the part that might be most immediately useful for everyday workers.

The model is integrated directly into Microsoft PowerPoint as a side panel, letting users create and edit presentations with Claude's help inside of PowerPoint instead of jumping between apps.

Claude in PowerPoint is available in research preview for Max, Team, and Enterprise plans.

What it means: Anthropic is evolving Opus from a coding powerhouse into a model that's useful for a much broader set of knowledge workers.

Product managers, financial analysts, and people across industries are already using it for non-coding tasks. This could be an inflection point for everyday business use.

Was this an intentional clapback? Probably.

OpenAI just released GPT 5.3 Codex less than an hour after Anthropic dropped Opus 4.6. It's their most advanced coding agent to date.

GPT 5.3 Codex set new industry records, scoring 77.3% on Terminal Bench 2.0. That's a 13 point leap over GPT 5.2 Codex and about 10 points ahead of Opus 4.6.

The model handles way more than coding though. Presentations, spreadsheets, debugging, deploying software, you name it.

OpenAI also describes it as the first model to play a key role in developing itself.

This came alongside the launch of a dedicated Codex Mac app for managing multiple AI coding agents simultaneously.

Here's the thing though.

You won't find GPT 5.3 Codex in ChatGPT. You have to use it in Codex.

And OpenAI's own Super Bowl ad was about Codex, not ChatGPT. Interesting shift.

What it means: OpenAI clearly means serious business on coding and agentic benchmarks.

The dual focus between ChatGPT and Codex could leave millions of potential users on the sidelines who don't think of themselves as coders but would absolutely benefit from this model.