Meta: Muse Models May Just Be the Spark That the Firm Needed in AI Model Development

Key Morningstar Metrics for Meta

• Fair Value Estimate
  : USD 850
• Morningstar Rating
  : ★★★★
• Morningstar Economic Moat Rating
  : Wide
• Morningstar Uncertainty Rating
  : High

On April 8, Meta META launched Muse Spark, a multimodal reasoning model with advanced tool use and agent orchestration capabilities. The model’s performance on various industry benchmarks was loosely in line with state-of-the-art models such as Anthropic’s Claude Opus 4.6 and Google’s Gemini 3.1.

Why it matters: A key bear case against Meta has been that, despite spending tens of billions of dollars on AI, the firm’s AI models have lagged those of leading-edge AI labs by a fair margin. We see Muse Spark as the first real model capable of competing with leading AI labs, a major win for Meta.

• We expect Meta to release more advanced versions in the Muse family of models, and we see value accruing to Meta in two key ways. First, we expect the firm’s data and ad targeting to improve, as Muse models can better analyze the troves of user data Meta collects for ad purposes.
• Secondly, with better AI features, including health-related ones, we expect Muse models to drive both sessions per day and time-spent-per-session higher, leading to more ad impressions and sales for Meta across its various platforms.

The bottom line: We maintain our USD 850 fair value estimate for wide-moat Meta. We continue to view shares as materially undervalued, even after accounting for the post-model-launch rally in share price.

Key stats: We noted Meta’s disclosure that Muse Spark is significantly more compute-efficient than Llama 4, the firm’s previous model, as well as open-source models by Chinese labs, including DeepSeek and Kimi.

• We think this improvement in efficiency is important in the context of digital ads, with Meta now being able to run agentic and reasoning-capable ad optimization and targeting at a lower marginal cost per ad served.