DeepSeek

DeepSeek

⚠️ Sample. An AI-generated demo of LLM Wiki Newsroom — the "open source AI" topic is just the example corpus.

Overview

DeepSeek is the AI company whose R1 model, released in January 2025 under the permissive MIT license, is a prominent OpenWeights example. DeepSeek published R1's weights and parameters along with a technical paper describing its Mixture-of-Experts architecture, but not the underlying training data — which is why R1 is classified as open weights rather than fully OpenSourceAI. R1 caused significant disruption on claims that it matched well-known LLMs while costing roughly $6 million to train, a fraction of comparable models. Because the weights are MIT-licensed and public, users can fine-tune R1 on their own data.

Connections

• OpenWeights — the release category R1 exemplifies
• ModelLicensing — R1's MIT-licensed weights
• FineTuning — enabled by the public weights
• OpenSourceAI — the stricter standard R1 does not fully meet