How much have major llm providers open sourced their models
Here is an overview of the extent to which OpenAI, Anthropic, Meta, and Le Mistral have open-sourced some or all of their models:
OpenAI
OpenAI has not open-sourced its models. Its flagship models, such as GPT-4 and GPT-3.5, are proprietary and only accessible via API. OpenAI has focused on providing these models as paid services through its platform, offering no public access to the underlying code or model weights. This approach contrasts with the open-source trend, as OpenAI prioritizes control over its models for commercial and safety reasons[8].
Anthropic
Anthropic, like OpenAI, has not open-sourced its models. The Claude family of models (including Claude 3.5) is available through APIs on platforms like AWS Bedrock and GCP Vertex AI but remains proprietary. Anthropic emphasizes safety and alignment in AI development, which may contribute to its decision to keep its models closed-source[2][6]. There is no indication that Anthropic plans to release model weights or training data for public use.
Meta
Meta has embraced open-source AI more fully than OpenAI or Anthropic. Its LLaMA series of large language models (LLaMA 2 and LLaMA 3) are available under various licenses, including open-weight releases for researchers and developers. Meta's LLaMA 3.1, the largest model with 405 billion parameters, is touted as the most powerful open-source model to date, with weights available for customization and fine-tuning[5][9]. However, there has been some criticism regarding whether Meta’s models are truly "open source" because the company does not release training data or algorithms[11]. Despite this, Meta’s commitment to open-weight models makes it a leader in the open-source AI space.
Le Mistral
Le Mistral is a strong advocate for open-source AI. The company has released several models with open weights under permissive licenses like Apache 2.0. Notable models include Mistral 7B and Mistral NeMo (12B parameters), both of which are fully open-source and available for modification and fine-tuning[4][7][10]. Le Mistral’s emphasis on transparency and decentralized development positions it as one of the most prominent players in the open-source AI movement.
Summary Table
| Company | Open-Sourcing Approach |