The opinions that guide participants in the market for investment in large language model development are not based on anything other than the views of other market participants. Mathematical models of such markets are not available in the literature. The proposed model assumes that an individual’s perspective can be adequately characterized by a metric of its veracity. Each interaction serves to converge the viewpoints of the interacting parties, with the level of trust placed in a counterparty as an information source being contingent solely on their integrity in upholding agreements. Transactions generate nominal added value, whereas real added value depends on nominal value and on the average truthfulness of agents’ opinions. The difference between these two quantities characterizes the scale of the emerging financial bubble, while their ratio serves as a measure of efficiency. Computer experiments were conducted to study the properties of the model. These experiments revealed four phases in the evolution of the market under study: trust accumulation, clustering, convergence, and stationary dynamics. They also showed that the presence of agents with fixed opinions does not prevent the convergence of opinions among the remaining participants. Both duration and outcome of convergence are shown to depend primarily on the initial distribution of opinions. The fact that optimists dominate the investment market for large language models is shown to be insufficient to claim the presence of a large-scale bubble. Directions are outlined for empirical testing of the model’s predictions (concerning, in particular, the dynamics of technical efficiency and its variance) and for further refinement of applied agent-oriented models.