Meta has introduced Muse Spark 1.1, the latest version of its AI reasoning model, calling it a major upgrade over the earlier Muse Spark. The company says the new model is designed for agentic AI tasks and comes with significant improvements in coding, computer use, multimodal understanding, and reasoning.
Announcing the launch, Meta said, "Today, we're excited to introduce Muse Spark 1.1, the latest model from Meta Superintelligence Labs and a significant upgrade from Muse Spark. Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with major gains in tool and computer use, coding, and multimodal understanding."
The company added that the new release moves it closer to its long-term goal of building personal superintelligence that can help people achieve everyday goals, create content, strengthen relationships, and complete important tasks.
Alongside the model, Meta has also launched a public preview of the new Meta Model API, allowing developers to build applications using Muse Spark 1.1. The model is now available in "Thinking" mode through the Meta AI app and Meta AI website.
Built for complex tasks
One of the biggest upgrades in Muse Spark 1.1 is its ability to work as an AI agent that can plan, organise, and complete tasks across different apps and services.
According to Meta, the model can create a plan, collect the required information, and even divide work among multiple AI sub-agents to complete tasks more efficiently. It can also adapt to new tools and custom skills without requiring additional training.
Meta said Muse Spark 1.1 can manage a context window of up to one million tokens, allowing it to remember earlier conversations, retrieve information from long interactions, and keep only the most important details for future tasks.
ALSO READ: Meta To Launch Custom AI Chip 'Iris' In September To Cut Nvidia, AMD Reliance
Smarter computer use
Meta says Muse Spark 1.1 performs better in computer-use workflows that involve several applications and changing information.
Instead of clicking through every step, the model decides when it is faster to automate a task using scripts and when direct interaction with the interface is more practical. The company says this makes it more efficient while working across long sessions with minimal human input.
Major coding improvements
Coding is another area where Meta says the new model has improved significantly.
According to the company, Muse Spark 1.1 can identify and fix complex bugs, add new features to large software projects, and even handle large-scale code migrations. It also performs better in creating web applications and solving end-to-end coding tasks than the previous version.
Meta said the model has been trained to work smoothly with popular AI coding workflows. It supports features such as planning mode, goal conditioning, sub-agent delegation, and context management.
The company also revealed that developers and researchers across Meta are already using Muse Spark 1.1 every day. It said the model has shown strong results on the company's internal coding benchmarks and performs competitively against other leading AI models.
Better with images, videos and audio
Apart from coding, Muse Spark 1.1 also brings improvements in multimodal understanding. It can process text, images, audio, and videos together while carrying information across long workflows.
Meta says this allows the model to generate visual-to-code outputs, create highly detailed captions for images and videos, and complete multimodal tasks that require both understanding and action.
The company added that these capabilities become especially useful when the AI needs to analyse visual or audio information before performing actions on a user's computer.
Safety checks before rollout
Meta said Muse Spark 1.1 underwent extensive safety testing before launch under its Advanced AI Scaling Framework.
According to the company, the model was evaluated across areas including cybersecurity, chemical and biological risks, and loss of control. It also claims the model offers stronger protection against jailbreak attempts, prompt injection attacks, and other adversarial techniques.
Meta further said these improvements help reduce hallucinations and make the model less likely to produce misleading responses.
ALSO READ: Meta Faces Potential $1.4 Trillion Penalty Over Facebook, Instagram Addiction Accusations
Now available for developers
Muse Spark 1.1 is now available through the new Meta Model API in public preview, giving developers early access to build AI-powered applications.
Meta believes the combination of long-context reasoning, coding abilities, multimodal understanding, and agentic capabilities makes Muse Spark 1.1 a strong foundation for building advanced AI assistants capable of handling large and complex workloads.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
