Meta released Muse Spark 1.3 on Wednesday, rolling the model into Muse Code and the Meta Model API, according to a Meta post and a Techmeme item citing Axios. Meta says the update improves performance across coding and agentic tasks, with Axios reporting that the model is being offered at the same price as its predecessor. The release is framed around practical use in longer-running work rather than a single benchmark headline. Meta says Muse Spark 1.3 is designed to sustain longer-horizon tasks by collaborating with users, managing multiple workflows inside one long thread, and using tools to build context from messy or conflicting sources. The company says the model can revise gaps in its own plan and track what it has learned before producing a final deliverable. Meta also says it trained the model across a range of harnesses so it can generalize across different agentic environments. In practice, the company describes a model that is intended to be more interactive: asking clarifying questions when prompts are ambiguous, calling for user input when stuck, and confirming before taking consequential actions. For teams using Muse Code, the most relevant claims are around workflow continuity and instruction following. Meta says Muse Spark 1.3 follows complex, long-form instructions more reliably than earlier Muse Spark models and is better at preserving detailed requirements across multi-step tasks. The company also says the model is better at keeping track of multiple requests in messy, single-threaded contexts, including when a user interrupts or redirects earlier work. The rollout is not complete in every mode. Meta says previously available reasoning modes are available today, while “max reasoning” will arrive later after additional safety testing. The supplied materials do not include a date for that later rollout. The commercial takeaway is limited but notable: Techmeme’s summary of Ina Fried’s Axios report says Muse Spark 1.3 comes at the same price as the prior version. The cluster does not provide the actual pricing, usage limits, benchmark scores, or details from Meta’s evaluation report, so the performance claims should be treated as Meta’s characterization until more independent testing is available. Who benefits: Developers using Muse Code or the Meta Model API are the named audience for the updated model. Teams experimenting with agentic coding workflows benefit most if the model’s claimed gains in long-thread task management prove reliable. Who's exposed: Competing coding-assistant and model API providers face another large platform emphasizing agentic software work. Users are exposed to uncertainty around the unreported details: the supplied items do not include benchmark scores, exact pricing, or a timeline for max reasoning.