Meta is rolling out Muse Spark 1.1, a new multimodal reasoning model from Meta Superintelligence Labs, alongside the public preview of the Meta Model API. The model is now available in Thinking mode in Meta AI, while developers can start using it via Meta’s new API surface. Meta says the release brings major gains in agentic workflows, computer use, coding, and multimodal understanding compared with the first Muse Spark model.
Muse Spark 1.1 is built for long, tool-heavy tasks. It can plan work, call tools, operate across external apps and services, use MCP servers and custom skills, and coordinate parallel subagents. Meta says the model can manage a 1-million-token context window, retrieve information from much earlier in a task, and compact the context so later steps retain critical details. In computer-use scenarios, the model can decide whether to write scripts, click through interfaces, or batch actions depending on the task.
For coding, Meta positions Muse Spark 1.1 as a large-codebase model for bug fixing, feature implementation, migrations, automated screenshots, debugging, and validation loops. The company says it has trained the model to support agentic coding setups with planning mode, goal conditioning, subagent delegation, and context compaction. The model also adds stronger multimodal workflows, including visual-to-code generation, image and video captioning, and tasks that require the model to inspect visual or audio inputs while acting on a user’s behalf. The new API marks a shift from the earlier Muse Spark rollout, which was limited to Meta AI and a private API preview. Meta’s evaluation report states that Muse Spark 1.1 extends access to external developers via an API that supports tool calling, function calling, and developer prompts. The report also compares it with Muse Spark 1.0 and notes that the new version achieves higher benchmark performance across multiple domains, especially cybersecurity and agentic coding, while employing additional mitigations before deployment.
Meta says the model was evaluated under its Advanced AI Scaling Framework before release. The evaluation report says that the unmitigated Muse Spark 1.1 reached a high-risk threshold in the chemical and biological, as well as cybersecurity, domains, but Meta applied multi-layered safeguards and assessed residual risk as moderate or lower before launch. The company also reported lower jailbreak attack success than Muse Spark 1.0 on StrongREJECT v2 and lower prompt-injection attack success on AgentDojo.
Early API partners quoted by Meta framed the release around agentic development and enterprise use. Replit CEO Amjad Masad pointed to the model’s million-token context, multimodal support, search with citations, structured output, parallel tool calling, and an OpenAI-compatible API package. Cline CEO Saoud Rizwan highlighted tool usage and pricing as relevant to scaled coding workloads, while Box VP of AI Products Yashodha Bhavnani said Muse Spark demonstrated enterprise capabilities competitive with frontier models in Box’s internal evaluations.