Meta raises the stakes in AI coding battle with Muse Spark 1.1 rollout

The next front in the AI race isn't another chatbot.

It's code.

For months, companies such as OpenAI, Anthropic and Google have been racing to build AI systems that do more than answer questions. They write software, fix bugs, migrate codebases and increasingly carry out tasks that once belonged to teams of engineers.

Now Meta wants a place in that conversation.

The company has unveiled Muse Spark 1.1, an upgraded version of its coding-focused AI model that it says is designed for complex software development and agentic workloads. At the same time, Meta is opening a public preview of its new Model API, giving US developers direct access to the model and denoting a shift toward commercial AI services rather than consumer chatbots alone.

The timing is hard to miss.

Over the past year, AI coding assistants have moved from experimental tools to products businesses are beginning to rely on. Anthropic's Claude has become a favourite among many developers, OpenAI continues to expand ChatGPT's coding capabilities, while Google has folded Gemini deeper into developer workflows. Meta, despite its open-source Llama models, has largely watched that contest from the sidelines.

Muse Spark 1.1 signals that it no longer intends to.

According to Meta, the model is built to tackle large software projects instead of isolated snippets of code. It can debug applications, assist with code migration, understand images, video and documents alongside text, and carry out multi-step tasks with limited human assistance. Those capabilities place it squarely in the growing market for so-called agentic AI—systems designed to complete entire workflows rather than respond to prompts one at a time.

Developers are another focus.

Meta is offering the model through its new API with free introductory credits before moving to usage-based pricing, an approach that follows strategies already adopted by rivals competing for enterprise customers. Analysts say the move shows a broader effort to build recurring AI revenue beyond Meta's advertising business.

The release also fits into a much bigger reshaping of Meta's AI strategy.

In recent months, Chief Executive Mark Zuckerberg has reorganised the company's AI operations under Meta Superintelligence Labs, recruited researchers from competing firms and accelerated development of new models after criticism that Meta had fallen behind the industry's fastest-moving rivals. Muse Spark first emerged from that effort in April. Version 1.1 is its first major update.

The launch follows closely on the heels of Meta's rollout of Muse Image, an AI model for generating and editing images across its apps. Together, the releases suggest the company is building a family of specialised models instead of relying on a single general-purpose assistant.

For developers, the immediate question is less about benchmarks than reliability.

The AI coding market has become crowded with tools that seek to generate software in seconds. What increasingly separates them is how well they can understand large codebases, recover from mistakes and complete longer, interconnected tasks without constant supervision.

That's the benchmark Meta is chasing.

Whether Muse Spark 1.1 can persuade developers to switch from established coding assistants will become clearer as more users begin testing the model. But one thing is already evident: the competition has moved well beyond chatbots. The battle is now over who can build the AI software engineer that businesses trust to do the work.