Meta Opens Muse Spark 1.1 to Developers as It Pushes Into AI Coding

Meta has begun offering its new Muse Spark 1.1 model to US developers through a public API preview, a move that puts the company’s latest AI system directly in front of software builders and strengthens its push to compete in coding-focused AI. The release matters because Meta is now trying to turn its recent model launches into real developer adoption, not just consumer visibility.

The company says Muse Spark 1.1 is a major upgrade over the first Muse Spark release from April, with stronger performance on coding tasks, better bug detection and repair, and broader support for agentic workflows that can span multiple applications. Meta is also positioning the model as multimodal, meaning it can work across images, video and documents in addition to text.

What Meta announced

Meta said Muse Spark 1.1 is now available in “Thinking” mode inside the Meta AI app and on the Meta AI website, while US developers can access it through the newly launched Meta Model API in public preview. The company is also handing out $20 in free credits to each new API account to encourage testing and experimentation.

This is the first time Meta has opened the model through a developer-facing API, after initially keeping Muse Spark available only inside Meta’s own products. Over time, the earlier model became part of the chatbots embedded in Instagram and WhatsApp, and it also powered the company’s smart glasses.

Why the new API preview matters

The API release is more than a product update. It signals Meta’s intention to become a serious platform provider in the AI market, rather than relying solely on consumer apps to showcase its models. By putting Muse Spark 1.1 into the hands of developers, Meta is inviting outside teams to build applications, coding tools and automated workflows on top of its technology.

That strategy is important because the AI race is increasingly being defined by who controls the most useful developer ecosystem. Companies such as OpenAI, Google and Anthropic have spent years shaping the market with APIs, while Meta has had to catch up after a slower start in foundation models.

How Muse Spark 1.1 is different from the first version

Meta describes the new release as a substantial leap from the first-generation Muse Spark model, based on feedback from developers who tested the earlier system. The company says the latest version is better at more complex programming work, including spotting subtle errors and correcting difficult bugs.

Meta also says Muse Spark 1.1 is better suited to agentic use cases, where an AI system does more than answer questions and instead performs tasks across several steps or tools. According to the company, that includes workflows involving multiple agents and a range of applications.

Another upgrade is native multimodal perception. In practical terms, Meta says the model can understand and reason across images, video and documents, which can make it more useful for software teams working with mixed input types rather than plain text alone.

How does this fit into Meta’s broader AI push?

Meta’s latest move comes as the company tries to justify the enormous spending it has directed toward narrowing the gap with the AI leaders. Over the past year, Meta has overhauled parts of its AI organization and brought in prominent hires as it tries to close in on the capabilities offered by rivals such as OpenAI, Google and Anthropic.

The company’s launch cadence also shows a broader effort to make Muse a real product family. Earlier this week, Meta introduced Muse Image, an image generation model that drew criticism because of its ability to use other people’s Instagram content in generated images. That controversy underscores how quickly Meta’s AI strategy is becoming both more ambitious and more scrutinized.

Meta says Muse Spark 1.1 represents a “step-change” from the original version, reflecting upgrades driven by developer feedback and aimed at tougher coding and workflow tasks.

In other words, the company is no longer treating Muse as a demonstration model alone. It is trying to make the system useful enough that developers will build around it, whether inside Meta’s own apps or in separate software products connected to the new API.

What developers can do with Muse Spark 1.1

Meta is targeting software developers who want an AI model that can help inside coding environments, automate multi-step processes and interpret mixed media inputs. The company’s messaging suggests Muse Spark 1.1 is intended for practical engineering work rather than only conversational use.

Potential use cases include:

• finding and fixing code bugs
• supporting agentic workflows across apps
• handling multimodal inputs such as screenshots or documents
• powering custom assistants inside developer tools
• building multi-agent systems for task automation

The public preview format means Meta can gather more feedback while limiting the rollout to a controlled audience. That approach also gives the company a chance to refine the model and the API before a broader release.

How does it compare with OpenAI, Google and Anthropic?

Meta’s challenge is not just building a strong model; it is proving that it belongs in the same league as the firms that already dominate developer mindshare. OpenAI, Google and Anthropic have established reputations for high-performing coding assistants, large language models and agent platforms. Meta is now trying to show that Muse can compete on practical utility, not only scale or reach.

That competition is likely to hinge on several factors: model quality, latency, pricing, ease of integration, and how well the system handles real-world coding tasks. Meta’s free-credit offer suggests the company wants developers to test those details for themselves.

Why coding performance is a strategic test

Coding has become one of the clearest benchmarks for frontier AI systems because developers can quickly judge whether a model saves time or creates more work. If Muse Spark 1.1 can reliably find bugs, repair them and coordinate tool-based workflows, it could become more attractive to engineering teams than a model that only produces polished text.

That makes the new release a key test of Meta’s AI ambitions. A model that performs well in coding can also support broader enterprise use, especially if it can reason over documents, images and video in the same environment.

Timeline of Muse Spark’s rollout

Meta’s launch strategy for Muse has moved quickly since the model first appeared in the spring. The sequence below shows how the company has expanded its reach in stages.

What this means for Meta’s AI strategy

Meta appears to be shifting from catch-up mode to ecosystem-building mode. Rather than keeping its strongest models locked inside its own apps, the company is now inviting outside developers to experiment, build and potentially standardize on Meta tooling.

That is a necessary step if Meta wants to transform its spending into long-term strategic value. Consumer distribution through Facebook, Instagram, WhatsApp and its smart glasses gives Meta a big audience, but APIs are what determine whether developers treat a company as an AI platform leader.

If Muse Spark 1.1 can attract meaningful usage, it may help Meta build momentum beyond social products and into software infrastructure. If not, the company risks adding another model to a highly competitive field without changing the balance of power.

What happens next?

The immediate next phase will be developer testing. Because the API is only in public preview for US developers, Meta can gather performance data, monitor adoption and adjust the product before expanding access.

Longer term, the question is whether Meta can turn Muse Spark into a credible coding model that developers trust for real work. Success would strengthen Meta’s case that its AI investments are paying off. Failure would leave the company still chasing the leaders in the market it is trying to redefine.

For now, the message is clear: Meta wants Muse Spark 1.1 to be judged not just as another chatbot model, but as a serious tool for coding, automation and multimodal AI development.

Key facts at a glance

Meta’s latest release suggests the company is serious about making Muse a developer-facing platform, and coding performance may become the clearest proof point yet for whether its AI comeback is working.