Meta just made its move on the AI coding market. On Thursday the company publicly launched Muse Spark 1.1, a multimodal model built for agentic coding, according to TechCrunch AI. It’s aimed squarely at the kind of tools OpenAI and Anthropic have been selling for a while now. And Meta is pricing it to pick a fight.
Here’s what Spark 1.1 actually does. Meta says it can handle multistep reasoning, manage digital workflows, and deploy new features inside enterprise systems. It’s built for the messy, real work: fixing bugs, running large code migrations, and orchestrating tasks across a range of external apps and services. “Muse Spark 1.1 delivers exceptional performance in personal agentic tasks that require planning and orchestration,” the company wrote in a blog post.
Why the price tag matters
The headline isn’t the model. It’s the cost.
Reuters reports Meta will charge $1.25 per million input tokens and $4.25 per million output tokens. That lands it right next to Anthropic’s Claude Haiku 4.5 and OpenAI’s GPT-5.6 Luna, if a touch above. Cost is one of the sharpest battlegrounds in AI right now, and Meta knows it. When enterprises are running huge agentic workloads, token pricing is the difference between a tool they pilot and a tool they scale.
So yes, Meta is late. Anthropic and OpenAI have shipped models like this for a good while. But late and cheap is still a threat, especially when the buyers are enterprises watching every dollar of their AI bill.
Zuckerberg breaks a three-year silence
Here’s the tell on how much Meta cares. CEO Mark Zuckerberg posted on X to promote Spark, his first post on the platform in three years, per TechCrunch AI. His last one was back in July 2023, around the Twitter-to-X rebrand.
Zuckerberg called Spark “a strong agentic and coding model at a very low price,” and said it was “strongest at agentic performance, tool use, and computer use.” He also dropped a line worth noting: “more to come soon.” Translation: this isn’t a one-off. Meta’s got more models loaded.
What stands out to me is the framing. Meta isn’t pitching Spark as the smartest model in the room. It’s pitching it as the workhorse that does real automation without wrecking your budget. That’s a deliberate lane, and it’s a smart one.
A brutal week for standing out
Spark didn’t drop in a quiet week. Not even close.
- Meta also unveiled a new AI image-generation model on Tuesday, Muse Image.
- SpaceXAI shipped a new version of Grok.
- OpenAI dropped a whole new family of models, GPT-5.6, on the same Thursday.
That’s four major releases from four heavyweights inside a few days. TechCrunch AI put it plainly: the competition is as healthy as ever, and anyone who wants to stand out has their work cut out for them.
What this means for you
If you build with AI, this is good news. More agentic coding options means more leverage on price and capability. Meta undercutting or matching the incumbents pushes everyone to sharpen their rates.
A few things to watch:
- Benchmarks under real load. Meta’s pitch is agentic performance. The proof is how Spark holds up on long migrations and multi-tool tasks, not demos.
- The next releases. Zuckerberg said more is coming. Expect Meta to keep filling out the Muse lineup fast.
- The price war. With four vendors crowding the same space, token costs are only heading one direction. Down.
Meta showing up doesn’t crown a winner. It just makes the fight louder. And for the people actually shipping software with these tools, a louder fight usually means better options and lower bills. You can find the full details at the original source.