The AI arms race just kicked into another gear. One day after SpaceXAI dropped Grok 4.5, Meta has fired back with its own model, Muse Spark 1.1. The announcement stood out for more than the tech itself, since Mark Zuckerberg broke a three-year silence on X to post it himself.
“Today we’re releasing Muse Spark 1.1, a strong agentic and coding model at a very low price,” Zuckerberg wrote. “It’s available through our new Meta Model API and in Meta AI.” In plain terms, “agentic” means the model can carry out multi-step tasks on its own rather than just answering questions.
In follow-up posts, he laid out what it’s built for. “Muse Spark 1.1 is strongest at agentic performance, tool use, and computer use,” Zuckerberg said. “It does well on long-running tasks with 1M token context window, can delegate execution to sub-agents running in parallel, and is trained to use computer interfaces on desktop, mobile, or browser.” Boiled down, that means it can juggle big tasks, remember a lot at once, split work across helper bots, and actually operate a phone, browser, or computer the way a person would.
“The Meta Model API allows developers to build using Muse Spark for the first time,” said Zuckerberg. “Our focus is on delivering strong agentic and multimodal models at very low cost. More to come soon.” This is the first time outside developers can build apps directly on top of Muse Spark.
On pricing, Meta listed it at $1.25 per million input tokens, $0.15 for cached input, and $4.25 for output. That comes in cheaper than Grok 4.5, which runs $2 for input and $6 for output.
The timing wasn’t lost on Elon Musk, who replied to Zuckerberg with a single word: “jinx.” His company SpaceXAI launched Grok 4.5 just a day earlier, also pitched as far cheaper than the competition, but Meta just one-upped it in price.
Early testing backed up the hype. Rayan from Vals AI, a firm that reviews AI models, said Meta was selling itself short by calling this a “minor” release. What surprised him most was the price. He found Muse Spark 1.1 costs about a tenth of what rivals like Fable and GPT 5.5 charge, cheap enough that he said it’s actually cheaper than running a free model on your own computers.
He also found the coding got much better than version 1.0, and the model ran noticeably faster than both Opus 4.8 and GPT 5.5. On top of that, this is the first time Meta has let outside developers connect to one of its models directly, and Rayan expects a lot of startups to quickly test it in place of what they use now. On one test that measures how well AI handles legal work, Grok 4.5 sat in first place for about a day before Muse Spark 1.1 jumped ahead of it. His main takeaway was that no single model is best at everything anymore, so there’s real value in picking the right one for each job.
For consumers, this price war is good news. As Meta and SpaceXAI keep undercutting each other, the cost of running AI keeps falling, which means the apps, assistants, and services you use should get cheaper, faster, and more capable. It also means more startups can afford to build AI tools, so expect a lot more options landing in your hands over the coming months. This AI arms race is only beginning, folks.
Want to see more of our stories on Google?
P.S. Want to keep this site truly independent? Support us by buying us a beer, treating us to a coffee, or shopping through Amazon here. Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent Canadian media!
