Read release histories as product decisions
Model releases are not all alike. A research demonstration, a consumer assistant, an API, open weights, a cloud deployment, and an enterprise control plane may share a name while carrying different access, pricing, data, and safety conditions. When evaluating any provider, separate the announcement date from the date a specific account, region, or deployment received the capability.
A reliable timeline records the primary announcement, exact model or product name, delivery method, license or terms, documented limits, representative tasks, known failures, price snapshot, and next review date. That record makes it possible to understand a change without turning a marketing statement into a procurement conclusion.
OpenAI, Anthropic, and Google
OpenAI’s path moved from a public research organization to a combination of frontier models, a developer API, and ChatGPT as a broad consumer and business interface. GPT-2’s staged-release discussion, GPT-3’s API availability, GPT-4, and later ChatGPT-era releases show why “announced” and “available to my organization” are separate facts. The practical question is always the current model documentation, account entitlement, data terms, and fallback plan.
Anthropic, founded in 2021, brought Claude to market with a persistent emphasis on controllability, Constitutional AI, model documentation, and responsible-scaling concepts. Its history is best read not only through model names, but through the relationship among API access, safety policy, evaluations, and product controls. Google combined Google Brain and DeepMind in 2023, then placed Gemini across research, consumer products, AI Studio, and Vertex AI. Its releases similarly require a distinction between a research presentation, a consumer feature, and a cloud API.
Kimi, DeepSeek, GLM, and MiniMax
Kimi became widely recognized through long-context assistant experiences: a reminder that a model company can build awareness around a concrete user task rather than an API-first identity. DeepSeek’s public repositories, technical reports, API, V3, and R1 releases made engineering efficiency, reasoning, and weight availability central to its public story. “Open” remains a term to inspect carefully: weights, code, reports, licenses, and commercial rights are not interchangeable.
GLM connects a research lineage, the ChatGLM open-source community, commercial model services, and application products. MiniMax has built across text, video, voice, music, and agent-oriented products, so its path cannot be summarized as one chatbot series. For both, compare the particular open repository, consumer offering, and commercial API rather than assuming a related name implies identical capabilities or terms.
Use the history to choose responsibly
No release timeline tells you which provider to buy. It tells you where to verify the underlying claims. Test the models available to your account against the same sanitized tasks, define a quality and cost threshold, review data handling and regional availability, and plan for changes. Model families evolve quickly; a small internal evidence log will remain more useful than a stale ranking.