Meta has officially unveiled 'Muse Spark 1.1', its first proprietary AI coding model, marking a major step in the company's efforts to challenge OpenAI, Anthropic and xAI in the rapidly expanding AI coding market.
The announcement was made by Meta CEO Mark Zuckerberg, who returned to X with his first post since July 2023, ending a three-year absence and highlighting the significance of the launch.
Muse Spark 1.1 represents a strategic shift for Meta. While the company has traditionally focused on open-source AI through its Llama family of models, Muse Spark is a paid proprietary model aimed at enterprises and software developers who increasingly rely on AI to automate coding tasks.
Meta takes on OpenAI and Anthropic
The AI coding market has become one of the most competitive segments in artificial intelligence, with products such as OpenAI's Codex, Anthropic's Claude Code, and SpaceXAI's Grok already serving developers.
With Muse Spark 1.1, Meta hopes to gain market share by combining strong coding performance with significantly lower pricing.
Announcing the model on X, Zuckerberg described Muse Spark 1.1 as "a strong agentic and coding model at a very low price."
The announcement comes as businesses continue to look for ways to reduce AI expenses. Companies such as Uber and Walmart have reportedly imposed limits on employees' AI token usage due to rising costs associated with large language models.
Pricing designed to attract developers
Muse Spark 1.1 is available through the Meta Model API, allowing developers to integrate it into applications, developer tools and enterprise software.
Meta is offering $20 (around Rs. 1,900) in free API credits to every new developer account. Once those credits are exhausted, pricing is set at:
The pricing is lower than that of several competing models.
For comparison:
xAI's Grok 4.5 costs $2 per million input tokens and $6 per million output tokens.
OpenAI's GPT-5.6 Terra costs $2.50 per input token and $15 per output token.
Anthropic's Claude Opus 4.8 charges $5 for input tokens and $25 for output tokens.
By offering lower prices, Meta hopes to encourage wider adoption among startups, enterprises, and software teams looking to reduce AI infrastructure costs.
How Muse Spark 1.1 works
Unlike traditional AI chatbots that answer programming questions, Muse Spark 1.1 is an agentic AI coding model designed to handle complete software engineering workflows.
Developers interact with the model through the Meta Model API, which can be integrated with integrated development environments (IDEs), enterprise applications, developer platforms, and automated software pipelines.
Users can provide natural-language prompts, software requirements, or existing source code. The model analyses the context, understands project structure and generates or modifies code accordingly.
Muse Spark 1.1 is capable of:
Writing code from natural language instructions.
Finding and fixing software bugs.
Refactoring existing code for better efficiency.
Explaining large codebases.
Generating documentation and unit tests.
Managing enterprise software workflows.
Deploying new product features.
Performing large-scale code migrations across programming languages and frameworks.
Rather than assisting with a single coding task, the model is built to plan and execute multiple interconnected development tasks with minimal human intervention, making it suitable for enterprise automation.
Built for AI agents
Muse Spark 1.1 has been developed by Meta Superintelligence Labs (MSL) under the leadership of Alexandr Wang.
According to Wang, coding capabilities are essential for building advanced AI agents that can independently complete complex tasks.
He explained that the model was trained not only to generate high-quality code but also to integrate seamlessly with the software development tools and testing frameworks already used by developers.
Meta believes this compatibility will help accelerate adoption without requiring companies to change their existing workflows.
Benchmark performance
While introducing Muse Spark 1.1, Zuckerberg shared benchmark comparisons showing the model outperforming Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 in several coding-related tasks.
However, the published benchmarks did not include comparisons with Anthropic's newer Mythos and Fable models or OpenAI's latest GPT-5.6 model.
Meta says Muse Spark 1.1 is among its strongest coding models yet and is specifically optimized for enterprise software engineering workloads.
Meta expands beyond open-source AI
The launch follows Meta's recent introduction of Muse Image, its latest AI image-generation model available through the Meta AI app, WhatsApp, and Instagram.
Muse Spark 1.1 also marks an important change in Meta's AI strategy. Unlike the company's open-source Llama models, Muse Spark is a proprietary commercial product that developers must pay to use through the Meta Model API.
Despite the shift, Alexandr Wang said Meta remains committed to open-source AI. He confirmed that an open-source version of Muse Spark is currently in development, although the company has not announced a release date.
Wang also revealed that Meta is already working on an even more advanced AI model internally, currently codenamed "Watermelon."
Meta intensifies the AI race
With Muse Spark 1.1, Meta is making its strongest move yet into the AI coding market. By combining enterprise-focused capabilities, autonomous coding features and aggressive pricing, the company aims to compete directly with OpenAI, Anthropic and SpaceXAI.
As demand for AI-powered software development continues to grow, Muse Spark 1.1 could become a key part of Meta's broader strategy to expand its presence in commercial AI services while balancing proprietary offerings with its long-term commitment to open-source AI.
