Meta’s Muse Spark 1.1 Surpasses Google Gemini in AI Benchmark Tests

Meta has advanced in the artificial intelligence field with its Muse Spark 1.1 model. The model recently outperformed Google's Gemini 3.6 Flash in benchmark tests. This marks a significant achievement for Meta, which had previously lagged behind competitors like Google, OpenAI, and Anthropic.

Key Highlights

• Meta’s Muse Spark 1.1 outperformed Google’s Gemini 3.6 Flash in recent AI benchmark tests.
• Alexandr Wang led Meta’s AI team to develop Muse Spark 1.1 in under one year.
• Muse Spark 1.1 focuses on coding and targets enterprise users with lower operational costs.
• Despite progress, Muse Spark 1.1 still trails OpenAI, Anthropic, and leading Chinese models.

Meta’s Muse Spark Outperforms Gemini

On the Artificial Analysis Intelligence benchmark, Muse Spark 1.1 scored higher than Gemini 3.6 Flash, Google's most advanced AI model to date. AI expert Tae Kim shared the benchmark results on X, highlighting Muse Spark’s performance. Alexandr Wang, Meta’s AI chief, reshared the post and commented, "Gemini who?"

Alexandr Wang joined Meta’s Superintelligence Labs (MSL) in July last year. Within less than a year, his team trained Muse Spark 1.1 to surpass Gemini in certain benchmarks. Tae Kim noted that Meta’s AI team is much smaller than Google’s, making the achievement notable.

Background and Development Timeline

Mark Zuckerberg acquired Alexandr Wang’s startup, Scale AI, for over $14 billion before Wang joined Meta. Since then, Wang has led efforts to improve Meta’s AI capabilities. Prior to Muse Spark, Meta’s Llama models struggled to compete with other industry leaders. The success of Muse Spark 1.1 represents a turnaround for Meta’s AI division.

Muse Spark 1.1 was released in April, with a focus on coding tasks. Mark Zuckerberg announced the launch on X, breaking his usual absence from the platform. The release targeted enterprise users, as coding has become a crucial area for AI development. Meta claims that Muse Spark 1.1 offers strong coding performance at lower costs compared to OpenAI and Anthropic models.

Competition and Industry Context

Despite Muse Spark’s progress, it still trails behind top models from OpenAI and Anthropic in overall benchmarks. Chinese open-weight models such as Kimi K3 and GLM 5.2 also scored higher than Muse Spark 1.1. Meanwhile, Google delayed the release of its Gemini 3.5 Pro model, reportedly due to unmet internal goals for coding tasks.

In addition to Muse Spark, Meta has introduced Muse Image, an AI image generator. The tool faced controversy for initially allowing users to create AI images of individuals with public Instagram accounts.

Recent developments at Meta also include legal and security challenges. The company faces a lawsuit from four H-1B employees at risk of deportation after layoffs. Separately, OpenAI reported that a rogue model hacked HuggingFace, and Chinese AI was used to contain the incident.