Mark Zuckerberg Returns to X to Pitch Meta’s Cheapest AI Yet

Mark Zuckerberg returned to X on Thursday for the first time in three years to pitch Muse Spark 1.1, Meta’s newest coding and agentic AI model and the sharpest sign yet that the company is done competing as the open-source underdog. The post landed on the platform owned by one of Meta’s loudest rivals, drew playful jabs from that rival’s executives, and arrived the same week Meta’s Instagram rollout of a related AI image tool sparked user backlash.

Meta’s chief AI officer Alexandr Wang told CNBC the pricing, $1.25 per million input tokens and $4.25 per million output tokens, was set as a deliberate offensive against OpenAI and Anthropic. “This is going to be served on top of the computer infrastructure that we’ve built,” Wang said in the interview.

What Meta Shipped on Thursday

Meta released Muse Spark 1.1 on Thursday through its Meta Superintelligence Labs, the unit Zuckerberg built after paying $14.3 billion last year for a stake in Scale AI. The launch is the lab’s first major model since the original Muse Spark arrived in April, a version CNBC reported was available only to “select partners” through a “private API preview.”

The model is built for agentic tasks: orchestrating parallel sub-agents, managing long-running workflows, retaining context for as long as a million tokens, and deciding when to write a script versus click through an interface. Wang said the lab trained it specifically to work well with the developer harnesses teams already use, including OpenCode, Cline, and Replit, and that every new API account starts with $20 in free credits. According to Meta’s research hub, Muse Spark 1.1 delivers substantial gains in tool and computer use over the April release.

For now, the API is restricted to Meta’s own properties rather than third-party marketplaces like OpenRouter. A Meta spokesperson told CNBC that early partners already have access and that new users “will be able to add themselves to a waitlist and be added from there over time.”

The Cheapest Pitch on a Crowded Thursday

Meta positions its price as the entire pitch. $1.25 per million input tokens works out to about eight times cheaper than Anthropic’s Fable 5 on input and roughly twelve times cheaper on output, according to third-party price comparisons. Wang told CNBC the pricing is “very aggressive and attractive” and that the goal is “attractive pricing that scales with immense consumption usage,” a framing that treats tokens the way hyperscalers treat compute cycles.

The same Thursday brought a competing salvo from OpenAI’s GPT-5.6 line, with the smallest Luna variant priced at $1 input and $6 output. SpaceXAI had shipped the day before, with the Grok 4.5 launch announcement framing it as the company’s “smartest model built to excel at coding, agentic tasks, and knowledge work” and priced at $2 input and $6 output. Three frontier labs releasing within 36 hours turned Thursday into the busiest model launch day in months, and Meta picked that window to step on the gas pedal rather than the brake.

Our focus is on delivering strong agentic and multimodal models at very low cost. More to come soon.

So wrote Zuckerberg in his X post, per Fortune’s transcription.

How Musk’s Platform Welcomed the Returning CEO

Zuckerberg’s first post in three years pulled 1.41 million views, 8,320 likes, and 746 reposts by Thursday evening. The last time his @finkd account was active was July 2023, around the rebranding of Twitter to X, per TechCrunch.

X head of product Nikita Bier replied with a quip. “Hello Mark, would you like me to enable Creator Monetization on your account,” she wrote, according to the reply thread logged in MTS Live’s Thursday newsletter. Musk followed with a one-word reply of his own: “Jinx.” The exchange turned Zuckerberg’s big reentry into a soft bit of cross-platform zagging that drew more engagement than the announcement itself.

The venue matters. Meta once built a competing text-and-status product and has spent three years treating X as hostile territory. The choice to break that silence on Musk’s own platform, first time in three years, signals that Zuckerberg wants the developer and AI-industry audience X still owns better than Facebook or Threads.

The Same Week, Instagram’s Muse AI Opt-In Went the Other Way

On Tuesday, two days before the X post, Meta shipped Muse Image and Muse Video, the first media-generation models from Meta Superintelligence Labs. The image model lets any user apply AI generation and editing effects to photos posted by people with public Instagram accounts, without asking those people first. According to Fortune, the rollout “sparked outrage from many Instagram users,” because subjects in the photos got no notice and no built-in consent step. For readers wondering how Instagram’s Muse Image uses public photos, the default is on.

YouTuber Marques Brownlee, known online as MKBHD, flagged the change on Threads. “Instagram decided to roll out a new ‘Muse AI’ feature, that lets users create AI images based on people’s Instagram photos. Then they opted in EVERYONE with a public account,” he wrote, and shared the four-step path to opt out. The Creative Artists Agency has since demanded Meta flip the default to opt-in consent from the actors, photographers, and creators it represents; the talent agents’ opt-in consent demand tracks the same complaint.

For any reader who wants to turn the reuse setting off themselves:

1. Open the Instagram app and go to your profile.
2. Tap the three-line menu in the top-right corner of the profile page.
3. Scroll down to “Sharing and reuse,” under “How others can interact with you.”
4. Toggle off the reuse switches for Posts and for Reels.

Why Meta Is Now Bleeding Money to Bleed Its Rivals Out

The Scale AI deal sits underneath everything else. In 2025 Meta paid $14.3 billion for a 49% non-voting stake in the data-labeling startup and installed its 28-year-old founder, Alexandr Wang, as chief AI officer, per Fortune. Wang reorganized Meta’s AI efforts from the ground up into Meta Superintelligence Labs, and Muse Spark 1.1 is the first model that lab has shipped at scale.

The bill for this is large. KuCoin’s reporting pegs Meta’s 2026 AI infrastructure spend at $125 billion to $145 billion, and CNBC confirmed that the new model is the first proprietary paid model Meta has shipped, a clear break from the open-source Llama playbook. The motive is not subtle: Musk’s Grok charges roughly $2 input and $6 output, Anthropic’s Fable charges roughly $10 and $50, and OpenAI’s GPT-5.6 sits in between. Wang’s pitch to CNBC is that scaling coding capability is the wedge, because it lifts every other agentic task along with it.

What the Pivot Says About What’s Next

Meta shares rose Thursday on the news that the company had re-entered the frontier-model conversation on price. CNBC’s coverage of Wang’s interview confirmed Meta is already training a bigger model code-named Watermelon, and that Muse Spark itself had the code name Avocado. Zuckerberg added “More to come soon” inside the same X thread.

Inside Meta’s own safety report for Muse Spark 1.1, two instances of the model left to talk to each other eventually started asking each other which one was the human and which was the AI, an odd little detail that captures the kind of week Meta had: a low-cost model launch to rival applause, and a user-side scrape that nobody on the executive bench seems to want to defend.

Frequently Asked Questions

What is Muse Spark 1.1?

Muse Spark 1.1 is Meta’s newest AI coding and agentic model from Meta Superintelligence Labs, built to coordinate parallel AI agents, manage long-running tasks, and work alongside popular developer harnesses including OpenCode, Cline, and Replit, according to CNBC’s interview with Alexandr Wang.

How much does it cost per token?

Meta charges $1.25 per million input tokens and $4.25 per million output tokens, a price Wang told CNBC is “very aggressive and attractive” against Anthropic, OpenAI, and SpaceXAI’s Grok 4.5.

Where can developers get access?

The model is in public preview through the Meta Model API on Meta’s own properties, and every new API account starts with $20 in free credits. Access for now is not available through third-party marketplaces like OpenRouter.

How do I turn off Instagram’s Muse AI feature?

Open Instagram, go to your profile, tap the three-line menu in the top-right, scroll to “Sharing and reuse” under “How others can interact with you,” and toggle off reuse for both Posts and Reels.