Meta Now Charges for Its AI: What Muse Spark 1.1 Means for You

If you have followed AI news even loosely, you know Meta as the company that gave its models away. Its Llama models were 'open weight' and free to run, and that was the whole pitch. So this week's news is a genuine turn: on 9 July 2026 Meta released a new model called Muse Spark 1.1, and for the first time it is charging developers to use its own model. Below, in plain English, is what the model is, what 'agentic coding' means, and what actually changes for you now that Meta has put a price on the door.

Muse Spark 1.1 is Meta's newest AI model, and Meta calls it its most capable yet for real-world coding and 'agentic' tasks. Think of it as the engine, not the app. On its own it is a large language model, the same broad kind of technology behind ChatGPT and Google's Gemini. What makes this version notable is where Meta is aiming it: at software developers who want an AI that can help write, fix and run code, and increasingly do so without a human holding its hand at every step.

One thing to keep in your pocket as you read: 'most capable yet for coding and agentic tasks' is Meta's own description, based on Meta's own testing. That does not make it false, but it is a maker's claim, not an independent scorecard, so treat it the way you would treat any company describing its own product. The independent comparisons usually come later.

'Agentic' is the buzzword doing the heavy lifting here, so let's unpack it. A normal AI assistant answers one thing at a time: you ask a question, it replies, you ask the next question. An agentic AI is trusted to take a goal and work through the steps by itself. In coding terms, instead of you saying 'write me this one function,' you might say 'find the bug that's breaking the login page,' and the AI reads the files, tries a fix, runs the code to see if it worked, and tries again if it didn't.

That is why every big lab is chasing it right now. Muse Spark 1.1 landed in one of the busiest launch stretches of the year, arriving alongside OpenAI's GPT-5.6, Google's Gemini 3.5 Pro and xAI's public Grok 4.5. All of them are competing on roughly the same promise: an AI that can reliably carry out multi-step coding work without going off the rails. So the real test for Muse Spark 1.1 is not whether it sounds clever, but whether it can be trusted to finish the job.

Here is the part that matters even if you never touch code. Meta is now charging developers to use Muse Spark 1.1, through a new product called the Meta Model API. An 'API' is simply the paid pipe that lets an app send your request to Meta's model and get an answer back, the same way OpenAI and Anthropic already sell access to their models. The shift is that Meta, the company famous for free, is now standing up a paid, hosted model business of its own.

So what is 'open weights,' and is the free option going away? 'Weights' are the model's finished settings, the giant table of numbers it learned during training. When a company releases them openly, anyone can download that file and run the model on their own computers, for free, without asking permission. That was Meta's Llama approach for years. The timing here matters, because rivals like DeepSeek, Meituan's LongCat and Alibaba's Qwen now give away strong open models too, which chips away at the very thing that made Meta's free strategy special. A paid API is one way Meta can start making money directly from every request instead.

If you are a casual reader, nothing changes tomorrow: you don't pay Meta to use AI, and the apps you already use are unaffected by this. If you build things, or you use tools built by small developers, this is the real headline: another paid AI supplier just opened for business, which is good for choice and competition but is one more meter running. And this all lands right after Meta's painful internal reorganization, in which roughly 8,000 roles were cut and about 7,000 people were moved onto AI teams, so Muse Spark 1.1 is being watched as an early sign of whether that upheaval produces anything worth paying for. The reader takeaway is simple: Meta is no longer just the 'free AI' company, and from here you'll want to check whether the AI inside a tool you rely on is the free open kind or the new paid kind, because that quietly shapes what it costs and how long it sticks around.