Meta Is Back in the LLM Game After a Year-Long Break

Key Takeaways

• Muse Spark released April 8, aiming to rival top LLMs
• Llama 4 faced criticism for benchmark fudging and poor performance
• Meta invested $14.3B and hired top talent, including $250M offer
• Avocado model delayed after internal tests lagged behind competitors

Pulse Analysis

Meta’s LLM journey has been a roller‑coaster. After the Llama 4 rollout in April 2025—promised as a multimodal breakthrough but widely condemned for inflated benchmark results and subpar real‑world performance—Meta’s reputation among researchers took a hit. The backlash highlighted a broader industry lesson: raw scores matter less than consistent, trustworthy outcomes. In response, Meta overhauled its AI organization, injecting $14.3 billion into infrastructure and talent, most notably an acquihire of Scale AI’s CEO and a $250 million compensation package to lure elite engineers.

The April 8 debut of Muse Spark marks the first public offering from Meta’s newly formed Superintelligence Labs. Accompanied by a 158‑page safety and preparedness report, the model posts impressive scores on standard benchmarks, leveraging Meta’s vast compute resources and proprietary data. Yet experts caution that benchmark dominance can mask gaps in reasoning, coding, and nuanced conversation—areas where post‑training fine‑tuning, a strength of Anthropic and OpenAI, proves decisive. Meta’s metrics‑driven culture may accelerate raw capability gains, but without comparable post‑training expertise, Muse Spark could struggle to achieve the “personality” that differentiates leading LLMs.

If Muse Spark meets or exceeds expectations, Meta could re‑enter the elite AI tier, pressuring rivals and reshaping talent dynamics. The company’s aggressive hiring spree—offering mid‑tens‑of‑millions salaries and multi‑hundred‑million packages—signals a long‑term commitment to out‑compete OpenAI and Google for top researchers. Conversely, the delayed Avocado model underscores the challenges of translating raw compute into market‑ready products. Stakeholders will watch closely whether Meta can convert its financial muscle into sustainable innovation, or if its resurgence will remain a series of high‑profile releases that fall short of industry standards.