Meta releases Muse Spark 1.1, claims edge over Google on coding and reasoning

Meta's release of Muse Spark 1.1 reflects a dramatic shift in the company's AI strategy and organizational structure. After long focusing on open-source AI development, Meta pivoted following its $14.3 billion（約2.3兆円） acquisition of a 49% non-voting stake in Scale AI in 2025. The company then reorganized its AI efforts under Alexandr Wang, newly appointed as its first chief AI officer, creating Meta Superintelligence Labs. The organizational changes were not smooth—staff reportedly felt whiplashed as Wang and Zuckerberg rebuilt the AI team from the ground up, including creation of an Applied AI unit in March that pulled in engineers for data collection tasks.

The timing and framing of Muse Spark 1.1's release underscore Meta's intent to compete directly with OpenAI and Anthropic. The model is proficient across coding, video captioning, and reasoning tasks, and benchmarks show it beats Google's latest Gemini on coding and reasoning capabilities. However, the company has faced credibility challenges—Meta confronted accusations that it manipulated benchmark results for a prior AI model release, with a former executive denying in April 2025 that the company trained on test sets. On concrete performance, Muse Spark 1.1 still trails the flagship models from Anthropic (Mythos 5, Fable 5) and OpenAI (GPT-5.6) on at least one coding metric, according to an open-source leaderboard.

Meta's commitment to aggressive pricing and low-cost deployment, paired with the recent launch of Muse Image and Muse Video models, positions the company as a volume player in the AI market. However, its decision to allow AI-powered image generation and editing effects on Instagram photos without explicit user permission sparked significant user backlash, suggesting that speed-to-market may outpace privacy and consent considerations in Meta's roadmap.