EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hugging Face Blog·Aug 14·5 min read
State of Open Models: Summer 2026 Observations
TL;DR
In 2026, the Hugging Face Hub saw a significant increase in public models and datasets, with Chinese labs dominating the landscape by releasing larger models than their American counterparts. This shift indicates a new trend where labs are skipping smaller models and focusing on high-parameter models directly.
✦ Why It Matters
Evaluate your current model deployment strategy to accommodate larger models from emerging labs.
Key Takeaways
- Explore Qwen for a broad range of model sizes and use cases.
- Consider optimizing models for specific hardware to enhance performance.
- Engage with community contributions to improve model adoption.
- Monitor the evolving landscape of open-source AI for strategic advantages.
Full Summary
The Hugging Face Hub saw significant growth in open models and datasets, with repositories increasing from 2.43 to 2.96 million. Chinese labs dominated the frontier, releasing models with up to 2.78 trillion parameters, while U.S. labs focused on optimizing existing models for their hardware.
Notably, AMD and NVIDIA led in new model releases, highlighting a shift from traditional model labs to hardware companies. The analysis revealed that community engagement is crucial, with Qwen emerging as a key player, generating over 151,000 derivatives.
This indicates a trend where model adoption is driven by community contributions rather than just new releases, reshaping the open-source AI landscape.