Language model agents in price competition
How deployment choices shape pricing behavior in repeated competition.
The problem
What happens when competing sellers delegate their prices to language model agents? In the homogeneous-goods Bertrand benchmark, sellers offer the same good and repeatedly adjust prices. The competitive benchmark has prices near marginal cost, but the behavior of learning agents depends on how they are deployed.
This project studies how the information agents receive, the actions they produce, and the frequency of their decisions shape the resulting price dynamics.
Raw-history deployment
Market histories including rival prices
One price each period
Prices can remain above the competitive benchmark
Profit-summary deployment
Summaries of the agent’s own profits
A price distribution at intervals
Prices tend toward the competitive region
What this work shows
- Agents choosing a single price each period from raw histories that include rival prices can sustain prices above the competitive benchmark in self-play.
- Agents periodically choosing price distributions from summaries of their own profits tend toward competitive prices in self-play.
- Mixing deployment interfaces can produce dynamic regimes with high-price anchors and repeated undercutting.
The central message is that deployment choices matter when evaluating competition between language model agents.
Publications
-
arXiv preprint
Bichler, M., Hierhager, P., Köck, K., & Pieroth, F. (2026). How language model agents reshape price competition. https://arxiv.org/abs/2610.21711Abstract
Repeated Bertrand competition is a stringent benchmark for algorithmic pricing: in homogeneous-goods markets, standard learning procedures typically select the stage-game Nash outcome with prices near marginal cost. We show that LLM pricing agents can escape this benchmark, depending on deployment decisions. LLMs outperform standard online learners through faster adaptation, but this advantage largely disappears against full-information learners. An extracted surrogate algorithm illustrates how LLMs in specific deployments profit from regularly undercutting slower learners. Agents that post one price every period from raw market histories that include rival prices can sustain supracompetitive regimes in self-play. Agents that periodically emit probability distributions from summarized histories of their own profits instead move toward the competitive region in self-play. Markets mixing these interfaces can generate dynamic regimes in which LLMs establish a high-price anchor and regularly undercut it to capture demand. These regimes can keep both competitors’ prices above marginal cost. Deployment choices shape these outcomes, shifting the focus from stationary equilibria to dynamic pricing regimes.
BibTeX citation
@misc{bichler2026language, title = {How language model agents reshape price competition}, author = {Bichler, Martin and Hierhager, Philip and K{\"o}ck, Kassian and Pieroth, Fabian}, year = {2026}, eprint = {2610.21711}, archiveprefix = {arXiv}, primaryclass = {cs.GT}, url = {https://arxiv.org/abs/2610.21711}, }