Research systems, models and enterprise integration
IBM traces its corporate history to the Computing-Tabulating-Recording Company created in 1911. Its AI record includes Arthur Samuel's checkers research, which studied learning through experience, and Deep Blue's 1997 chess match, which relied on specialized parallel search. These are distinct approaches rather than successive versions of today's language models. [1][2][3]
Watson's 2011 Jeopardy victory demonstrated natural-language question answering through IBM's DeepQA system. The computer interpreted clues and selected answers without an internet connection during the match. That historical demonstration should be distinguished from the later enterprise products carrying the Watson name. [4]
IBM also develops language models. Its Granite 3.0 release included 2B and 8B language models, safety models and mixture-of-experts variants under Apache 2.0. Its Bamba research combined attention and state-space layers with university collaborators and inference work involving Red Hat. The April 2025 Bamba article reports on an already released model; its publication does not establish the first release day. [5][6]
Model development coexists with integration work. IBM's August 2026 OpenAI partnership described bringing OpenAI models and products into Consulting Advantage and industry-specific client workflows, with further training and deployment work planned. It does not imply that IBM stopped developing its own models or independently prove the projected business benefits. [7]