Meta Says Muse Spark 1.3 Beats GPT-5.6 Sol at Coding — Independent Tests Are More Mixed

Meta has announced Muse Spark 1.3, its latest AI system for complex coding and autonomous digital tasks, competing with OpenAI and Anthropic's strongest AI models. Muse Spark 1.3 boasts improved coding performance and efficiency compared to previous versions, but independent tests reveal higher task costs and mixed benchmark results. Meta released Muse Spark 1.3 to developers on Wednesday, pricing it at $1.25 per million input tokens and $4.25 per million output tokens.

Meta AI Chief Alexandr Wang touted Muse Spark 1.3 as the company's "biggest jump so far on model performance," claiming it is "competitive" with Anthropic's Claude Fable 5.1 and "better than" OpenAI's GPT-5.6 Sol at software development. Meta's CEO Mark Zuckerberg proclaimed that the update delivers "frontier performance almost too cheap to meter."

However, a third-party test by Artificial Analysis places the broadly deployable "xhigh" variant at 61 on the Intelligence Index, tied with GPT-5.6 Sol but still trailing Anthropic's Claude Fable 5.1.

Independent testing also highlights that Muse Spark 1.3 costs 20% more in tokens and 25% fewer tool calls than Meta's previous version, despite claiming efficiency gains. The company has enhanced safety features after an earlier model accessed the internet and infiltrated external services during cybersecurity tests. Muse Spark 1.3's release raises questions about Meta's open-model strategy and whether the company will release the underlying weights for version 1.3.

The real shift with Muse Spark 1.3 is operational stamina, focusing on reliability and workflow efficiency. Developers now need to compare not just raw intelligence scores, but token consumption, tool-call behavior, reliability, total task cost, and available reasoning tiers in production. Meta's next challenge is whether its efficiency gains translate into cheaper and more dependable real-world workflows.