DeepSeek Narrowed the Perceived US-China AI Gap
DeepSeek’s model releases strengthened claims that Chinese laboratories had moved much closer to US leaders through efficiency and open development.
Share
Posts by tag
Open Models
10 posts
UAE’s Falcon 2 Pushes Open AI Models Into the Global Race
Abu Dhabi’s Technology Innovation Institute released text and vision-language Falcon 2 models as the UAE expands its sovereign AI strategy.
Share
Microsoft’s Phi-3 Makes the Case for Smaller Open Models
Phi-3-mini pairs a compact 3.8-billion-parameter design with curated synthetic data for capable on-device and lower-cost AI.
Share
Meta’s Llama 3 Raises the Bar for Open Models
Llama 3’s initial 8B and 70B releases combine stronger evaluations, broad distribution and new safety tooling as Meta builds an open AI ecosystem.
Share
Mixtral 8x22B Pushes Sparse Open Models Forward
Mistral AI’s Mixtral 8x22B activates only part of its 141-billion-parameter architecture, pairing open access with multilingual, coding and long-context capabilities.
Share
Stable LM 2 Expands Open Models Across 1.6B and 12B Sizes
Stability AI’s expanded Stable LM 2 family gives developers multilingual open models at two distinct scales, with the 12B version targeting RAG, tools and function calling.
Share
Databricks DBRX Raises the Bar for Enterprise Open Models
DBRX combines a 132B sparse mixture-of-experts architecture with curated training data and Databricks tooling for custom, governed enterprise AI.
Share
xAI Opens Grok-1’s 314B Mixture-of-Experts Model
Grok-1’s base weights and architecture are available under Apache 2.0, but the release excludes its training data and real-time data connections.
Share
Llama 2 Opens Commercial AI Development Beyond Closed Models
Meta’s Llama 2 family combines downloadable model weights with a commercial licence, expanding access while intensifying debates about openness and misuse.
Share
Meta’s Open Speech Models Expand Recognition to 1,100 Languages
Meta’s Massively Multilingual Speech project uses self-supervised learning and public recordings to extend speech recognition and synthesis to more than 1,100 languages.
Share