By Ken Muthomi - Reporter
United States, 10 July 2026 - Meta has announced the launch of Muse Spark 1.1, which is an upgraded AI model from the Meta Super Intelligence Labs, capable of multimodal reasoning, coding, computer use, and agentic capabilities.
In a statement, the tech giant said that the new model builds on the original Muse Spark and is engineered to accomplish tasks by planning, reasoning, and coordinating actions across multiple applications and services, with minimal human input.
According to Meta, the new model can function as both a primary AI agent and a subagent, meaning that it can break large projects into smaller tasks, assign work across parallel agents and combine the results, and also supports a context window of up to one million tokens.
The tech giant says that the model is better equipped to perform computer-use tasks spanning multiple applications.
Instead of relying only on graphical interfaces, Muse Spark 1.1 can determine when it is more efficient to automate tasks using scripts or interact directly with applications, enabling it to complete workflows faster.
According to the company, Muse Spark 1.1 is now available in Thinking mode through the Meta AI app and Meta.ai. The company has also launched a public preview of its new Meta Model API, giving developers their first opportunity to build applications using the latest AI model.
“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 stated.
“Early partners of Muse Spark 1.1 praise the model as a complete agentic foundation, pairing long context handling with strong coding and reasoning capabilities to handle large-scale agentic workloads,” it added.
Related articles
The model introduces stronger multimodal capabilities, enabling it to understand and reason across text, images, and audio, and can also generate code from visual inputs, produce detailed image and video descriptions, among others.
Meta has also highlighted major improvements in coding, noting the model can diagnose and fix complex software bugs, implement new features in enterprise-scale codebases and carry out large code migrations.
The model has also been optimised for an agentic coding environment, supporting features such as planning mode, goal conditioning and delegation of subagents.
Following evaluations, Meta has further said that the model is operating within acceptable risk thresholds across areas including cybersecurity, chemical and biological risks, and loss of control.
The company specifically noted that the model has shown better adversarial robustness, lower hallucination rates, and reduced sycophancy, compared to other models.
“Muse Spark 1.1 demonstrates strong resistance to direct jailbreaks and indirect attacks from untrusted data, prompt injection, and developer-prompt attacks,” Meta said.