Meta Abandons Open-Source AI, Launches Closed-Source Muse Spark Model
Meta spent $14.3 billion to acquire AI startup Scale AI and its CEO Alexandr Wang, who now leads Meta's new AI division. Meta's new Muse Spark model is closed-source, in contrast to their previous open-source Llama model. Benchmarks show Muse Spark excels in medical and scientific AI but lags in coding and abstract reasoning tasks.
Why it matters
Meta's shift from open-source to closed-source AI models signals a major strategic change and could impact the broader AI ecosystem.
Key Points
- 1Meta pivoted from open-source AI to a closed-source model with Muse Spark
- 2Muse Spark outperforms on medical and scientific AI benchmarks but trails in coding and abstract reasoning
- 3Meta's open-source Llama model was overtaken by Chinese open-source models like Qwen
- 4Meta's $14.3 billion acquisition of Scale AI was to bring on CEO Alexandr Wang as Meta's first Chief AI Officer
Details
In 2024, Mark Zuckerberg published a manifesto declaring open-source AI as the way forward. However, 18 months later, Meta released their first closed-source AI model, Muse Spark, which is locked behind an API with no public weights. Meta paid $14.3 billion for a 49% stake in Scale AI, primarily to bring on co-founder and CEO Alexandr Wang as Meta's first Chief AI Officer. Muse Spark scores well on medical and scientific AI benchmarks, but trails leading models like GPT-5.4 and Gemini 3.1 Pro on coding and abstract reasoning tasks. The article suggests Meta closed off Muse Spark due to the rapid growth of Chinese open-source models like Alibaba's Qwen, which overtook Meta's own open-source Llama model in popularity.
No comments yet
Be the first to comment