Loading…
An AI-assisted newsroom with a human journalist always in the loop.
Concepts
A model-release pattern where model weights are made available under a license, without necessarily making the full system open source.
The distinction matters because open weights, open source, research access, and commercial restrictions are often blurred in AI coverage.
Each fact carries its own source note. Draft and source-check labels mean the item should not be treated as a finished reference entry.
DeepSeek described DeepSeek-V3 as a mixture-of-experts language model with model checkpoints available through its linked repository.[2]
DeepSeek introduced DeepSeek-V3 on December 26, 2024.[2]
Hugging Face describes the Hub as a large open-source collection of models, datasets, and demos.[3]
Meta AI introduced LLaMA on February 24, 2023.[1]
Meta and Microsoft introduced Llama 2 on July 18, 2023.[4]
Meta described LLaMA as a collection of foundation language models ranging up to 65 billion parameters.[1]
Meta said Llama 2 was available free of charge for research and commercial use.[4]
Mistral AI described Mixtral 8x7B as a sparse mixture-of-experts model with open weights.[5]
Mistral AI released Mixtral 8x7B on December 11, 2023.[5]
Published 2023-02-24. Retrieved 2026-06-06.
Official Meta announcement of LLaMA for research access, including 7B, 13B, 33B, and 65B parameter sizes.
Published 2024-12-26. Retrieved 2026-06-07.
Official DeepSeek announcement linking to the DeepSeek-V3 model and technical report.
Published 2026-06-07. Retrieved 2026-06-07.
Official documentation for the Hugging Face Hub. Publication date is the retrieval date because the docs page does not expose a stable article date.
Published 2023-07-18. Retrieved 2026-06-07.
Official Meta announcement for Llama 2.
Published 2023-12-11. Retrieved 2026-06-07.
Official Mistral AI announcement for Mixtral 8x7B.