Meta on Thursday rolled out Muse Spark 1.1, a major update to its AI platform, just three months after launching its first model under AI chief Alexandr Wang. The launch appears to be a sign of just how urgently Meta Platforms CEO Mark Zuckerberg wants to close the gap with OpenAI and Anthropic.Wang described Muse Spark 1.1 in a recent CNBC interview as the lab’s “strongest model for agentic and coding work yet.” The first Muse Spark, internally called Avocado, shipped in April under a closed partner program with no public-facing API. This time around, Meta is opening a developer portal with a public waitlist. Don’t expect to find it on OpenRouter or other third-party marketplaces, though — Meta is keeping distribution on its own servers for now.It’s the second Muse launch in a matter of days. On Tuesday, the company released Muse Image, originally code-named Mango, an image-generation model designed to attract creators and advertisers.Undercutting rivals on priceIn a crowded market for AI tools, Zuckerberg is rethinking price. Muse Spark 1.1 ships with a paid developer tier — the first time Meta has put a price on any of its AI models, and a new revenue line investors have been waiting for.“Since this is not an open-source model, this is, I think, the first time that we’re doing a real serious API,” Zuckerberg said on the Bloomberg Intelligence podcast. “And the pricing is going to be very aggressive and attractive.”The new Meta Model API comes in at about a quarter of what OpenAI and Anthropic charge for their top-tier models. New accounts get $20 in free credits upfront. After that, it’s $1.25 per million input tokens and $4.25 per million output tokens.The pricing is only part of the shift. Until now, developers wanting to build with Meta’s frontier AI generally downloaded Llama weights and hosted the models themselves or relied on third-party cloud providers. Muse Spark 1.1 changes that equation by allowing developers to call Meta’s own hosted infrastructure via the new Meta Model API, putting the company in direct competition with OpenAI, Anthropic, and Google for API workloads.That represents a notable change in Meta’s enterprise strategy. Rather than treating downloadable models as the primary product, Meta is beginning to position its own platform as the destination for developers who want managed inference, lower operational overhead, and access to the company’s latest frontier models as soon as they’re released.“The pricing from some of the other labs is very extreme and has very high margins,” Zuckerberg noted. “We think that there’s a real ability to be able to offer frontier or very high-level intelligence at a much more affordable cost.”“The pricing from some of the other labs is very extreme and has very high margins.”Agentic AI takes center stageThe new model’s standout improvement lies in its agentic capabilities — systems that can autonomously complete multistep tasks on a user’s behalf. Zuckerberg described Muse Spark 1.1 as having “state-of-the-art or very close to it” agentic reasoning and tool use.Wang’s team at Meta Superintelligence Labs leaned heavily into coding performance, reasoning that a model fluent in code can juggle parallel workstreams and chain tools together — the core competencies an AI agent needs.For engineering teams, that focus reflects how enterprise AI development is evolving. Production agents increasingly spend less time generating text and more time calling APIs, writing code, coordinating workflows, and interacting with external tools. Optimizing for those capabilities makes Muse Spark 1.1 more relevant for developers building automation systems than those simply looking for another chatbot.“You kind of have to build coding capabilities as part of that in service of overall agentic capabilities,” Wang told CNBC. According to Wang, the model beat out competitors on benchmarks that test how well an AI works with third-party dev tools — and MSL deliberately optimized it for the harnesses developers are already using, betting that compatibility would drive adoption faster than raw benchmark scores.“You kind of have to build coding capabilities as part of that in service of overall agentic capabilities.”Zuckerberg also said the model outperformed Google’s Gemini on several internal benchmarks covering agentic workflows, coding, and multimodal reasoning. “That is a pretty interesting milestone because I think this may be the first time, at least that I can remember, that Meta’s models are better than all of the Google models,” he said.“That is a pretty interesting milestone because I think this may be the first time, at least that I can remember, that Meta’s models are better than all of the Google models.”Meta shelves open-source playbookLlama’s spring 2025 release was arguably a disappointment, and Zuckerberg responded by essentially starting over. He poached Wang from Scale AI to run it, cut staff, and reshuffled teams.The shift does not necessarily mean Meta is abandoning open models. Instead, it suggests the company now sees hosted AI services as an equally important business. Open-weight models helped establish Meta’s credibility with developers, but hosted APIs generate recurring revenue and keep developers within Meta’s ecosystem rather than sending inference workloads to cloud providers or competing AI platforms. The planned open-source version of Muse Spark suggests Meta is attempting to balance both approaches rather than choosing one over the other.Monetizing the AI investmentMeta’s 2026 capex projections are at record levels; the company is pouring hundreds of billions into compute, chips, and data centers, with a $10 billion facility in Canada as the latest addition. Wall Street has applied heavy pressure on Zuckerberg to show returns; following Meta’s April earnings report, shares fell nearly 9% despite strong revenue, driven by concerns over the lack of a clear AI monetization plan.The Muse Spark API launch, alongside plans for a consumer chatbot subscription and potential enterprise AI agents, serves as Meta’s direct response.Zuckerberg dismissed the idea that AI models will inevitably become indistinguishable commodities, pointing to Anthropic’s latest model, Mythos, as an example of companies gatekeeping technology. “The capabilities are not actually getting diffused or made broadly available to everyone,” he warned.For now, Meta hopes Muse Spark 1.1 — and its aggressive price point — will ensure high-quality intelligence remains accessible, powering free tools like the Meta AI chatbot while carving out a lucrative space in the developer market.The post Meta debuts Muse Spark 1.1 and it isn’t free appeared first on The New Stack.