The AI Model Tier List

The AI model landscape is rapidly evolving, emphasizing model efficiency and architecture. Hugging Face is reportedly seeking a $13 billion exit and has engaged an investment bank to explore offers, boasting over 2 million models, 1.5 million datasets, and 1.5 million AI applications. The narrowing gap between open models and frontier models is facilitating the integration of open models into serious business workflows.

NVIDIA is pursuing a non-exclusive licensing deal with Poolside valued at $6 billion, alongside a billion-dollar equity investment. This arrangement allows Poolside's founders to retain their positions, as it is neither an acquisition nor an Aquahire. Should NVIDIA acquire Hugging Face, they may release an open weight model capable of outperforming competitors by year-end. NVIDIA's increasing focus on open models is part of a broader competitive strategy, coinciding with a price increase for top-end chips by up to 17%, likely driven by rising memory costs.

Alibaba has raised $10 billion in a record-breaking share sale but may need more capital than initially expected for its AI investments. Michael Burry has pointed out Alibaba's significant progress in the low-cost LLM market in the U.S., although he remains skeptical about the company's share issuances and predicts a decline in its return on invested capital. The Chinese market is witnessing smaller companies going public with larger returns than those in the U.S., as evidenced by Unitry's debut, which highlights a strong appetite for China's embodied AI sector.

Theo's AI model tier list has generated substantial discussion, ranking Fable 5 in S tier and GPT56 Sol in A tier, while several other models occupy lower tiers. Despite its high ranking, Fable 5 has seen limited sales compared to Opus 4.8, which continues to dominate the market, partly due to Fable 5's 30-day data retention policy that deters enterprise adoption. AT&T plans to significantly increase its use of open models for AI tasks, finding them comparable or superior to previous generation models.

The AI model landscape is increasingly characterized by the use of different models for various tasks, reflecting inherent trade-offs in design. Open source AI is gaining market share from closed models, which is viewed as advantageous for AI infrastructure demand. Predictions indicate that the AI model market will bifurcate into three categories: cheap generalist models, state-of-the-art generalist models, and state-of-the-art specialist models. Enterprises are likely to experiment with various approaches to uncover new efficiencies in their operations.