Zuckerberg breaks a 3-year X silence to start an AI price war

On July 9, Mark Zuckerberg posted on X for the first time in more than three years. His last post there was in July 2023, back when he and Elon Musk were trading cage-match challenges. He came back for one reason: to launch an AI model on his rival's home turf. @finkd announced Muse Spark 1.1, calling it "a strong agentic and coding model at a very low price," available through the new Meta Model API and in Meta AI.

Sit with that choice for a second. Meta owns Threads, its own X competitor. Zuckerberg spent three years pointedly not posting on Musk's platform. But the developers he needs live on X, so that is where he showed up. When a CEO swallows that much pride for a launch, the launch matters to him.

Meta finally sells access to its models

The bigger news is buried in the follow-up. In a second post, @finkd said the Meta Model API "allows developers to build using Muse Spark for the first time," with a focus on "strong agentic and multimodal models at very low cost." For years Meta gave its models away as open weights and never charged for API access. That era just ended. This is Meta's first paid developer API, in public preview for US developers with $20 in free credits to start.

The model itself is a serious release: a 1 million token context window, training for computer use across desktop, browser, and mobile, and the ability to delegate work to sub-agents running in parallel. Alexandr Wang, who leads Meta's superintelligence effort, called it "our strongest model yet for agentic and coding work" in his own launch post, @alexandr_wang.

But the headline is the price: $1.25 per million input tokens and $4.25 per million output, with cached input discounted to $0.15. That is roughly a quarter of what the frontier tiers from OpenAI and Anthropic cost, from a company that can subsidize the fight with ad revenue indefinitely.

How good is it, actually?

Independent benchmarking firm Artificial Analysis published its numbers two days later. @ArtificialAnlys scored Muse Spark 1.1 at 51 on its Intelligence Index, an 8 point jump over Muse Spark 1.0 in just three months. That puts it in a tie with GPT-5.6 Luna and GLM-5.2, but clearly behind the frontier: Claude Fable 5 sits at 60 and GPT-5.6 Sol at 59. On cost per task, though, it came out near the very top of the field.

So no, it is not the smartest model available. It does not have to be. Meta is not trying to win the intelligence crown this round. It is trying to reset the price floor for "good enough" agentic AI. And since the API reportedly follows the OpenAI-compatible format most tools already speak, swapping it into an existing product is closer to changing a config line than rebuilding an integration.

Our honest read: this is the moment the AI market starts behaving like the cloud market. Once a giant with unrelated revenue starts underpricing compute-heavy services, everyone else has to respond, and the customer wins. Expect price cuts, cache discounts, and cheap mid-tier models from every lab within months.

What falling model prices mean for your business

If you run a small or mid-sized company, this news is better for you than any benchmark chart. The tasks most businesses actually want automated, answering leads, drafting quotes, chasing invoices, summarizing calls, do not need the smartest model on earth. They need a reliable mid-tier model wired into your workflow, and the price of exactly that just dropped hard.

It also means the durable investment is not the model. Models will keep leapfrogging each other every quarter. The thing that compounds is the workflow built around them: the intake forms, the follow-up sequences, the handoffs. Build that once, and you can swap the engine underneath whenever a cheaper or better one ships.

That is the layer we work at. New Face Design builds automation for trades, dental practices, and property managers across the Fox Valley, and we design every system so the AI underneath is replaceable. Our free process audit will show you, in plain English, which of your processes would benefit first.

Zuckerberg ending a three-year silence to post a price list tells you where this race is heading. The models are becoming interchangeable. The advantage goes to whoever puts them to work first.