Meta Set To Launch Upgraded Muse Spark AI Model With Stronger Coding Capabilities

Meta is preparing to launch an updated version of its flagship Muse Spark artificial intelligence model, with Chief AI Officer Alexandr Wang confirming that the release is expected "soon."

According to a report by Business Insider, the upcoming model performs on par with OpenAI's GPT-5.5 across several widely followed AI benchmarks, although it did not specify which tests were used.

Wang also indicated that the new version delivers significant improvements in coding and AI agent capabilities compared with the current Muse Spark model.

When released, it could position Meta against AI coding platforms such as Anthropic's Claude Code, which offers developer-focused features including tool integrations, desktop applications, and workflow automation.

The existing Muse Spark includes a "contemplating mode" that leverages AI agents to improve the quality of responses. During internal testing, Meta found that enabling the feature improved the model's score on the HLE benchmark by 8%.

In response to a question on X about matching the coding performance of Anthropic's Claude Opus 4.8, Wang replied that Meta would achieve that goal "pretty soon."

The report suggests the improved model achieves its higher performance by using substantially more computing resources—around an order of magnitude more than the current Muse Spark model.

Meta's continued focus on coding capabilities signals broader ambitions in AI software development. According to a recent Bloomberg report, the company is considering launching an AI infrastructure service that could provide developers with either hosted AI models or raw computing capacity.

Beyond software development, Meta is also expected to expand Muse Spark's use cases into broader enterprise productivity, where AI models are increasingly being deployed to support business functions across industries such as healthcare, finance, and customer service.