Same architecture, different paths
After the transformer breakthrough (see "The transformer: the paper that changed everything"), several research labs and companies used the same base architecture - but with different goals, training data, and business models. That's how today's, at first glance confusing, vendor landscape emerged.
OpenAI: from non-profit to commercial front-runner
OpenAI started in 2015 as a non-profit organization with the stated goal of developing safe AI for the benefit of humanity. With the commercial success of GPT-3 and ChatGPT, the organization shifted increasingly toward a commercial product - access mostly only via API/subscription, the model itself staying proprietary (see "Open source vs. proprietary AI models: what's the difference?").
Google/DeepMind and Anthropic: research heritage and safety focus
Google/DeepMind co-developed the transformer architecture itself and brings that research depth into its own models (Gemini). Anthropic was founded in 2021 by former OpenAI employees, with an explicit focus on AI safety as a founding motive - both stay with proprietary models accessible via API.
Meta: the open counter-movement
Meta chose a different path: with the Llama model series, the company releases the model weights openly - anyone can download and run them themselves (see "Open source vs. proprietary AI models: what's the difference?"). That's not charity, it's a strategic decision: open models foster an ecosystem where many developers build on Meta's technology.
Why this matters for you as a decision-maker
If you understand which founding history and strategy a vendor comes from, marketing claims and roadmap decisions become easier to read - a safety-focused startup communicates differently from a commercial market leader or an open ecosystem play. That's not a value judgment, but a tool for reading vendor pitches (see "See through vendor pitches") in context.