WiremarketsBrief
In Controlled LLM Markets, Full Disclosure Helped Some Models and Hurt Others
Experimental results across trade, bargaining and auctions show that disclosure effects vary by model and market structure. Builders need to test information policy alongside model choice.
Brief
Researchers posted a preprint dated November 5, 2026, testing how much detail large language models receive before bargaining or trading. Ten models were run through three settings covering multiple resources, one-on-one bargaining and double auctions, with incentives and rules held fixed. Results were mixed. GPT-6 Sol's realized gains slipped from 97.7% to 90.8% when it moved from coarse to full disclosure in a two-resource trade test, whereas Claude Sonnet 5 climbed from 72.1% to 85.6%. In double auctions, Claude Opus 5.5's realized gains fell from 94.6% to 64.1% under full disclosure even though its candidate quality stayed above 0.99. The authors say the gap is not explained by a general mechanism and caution that the findings cover selected controlled markets, not live use.
Why it matters: For developers building multi-agent systems, the findings suggest disclosure choices can shift outcomes and that those effects differ by model.
Read the full story
Dextape context
Built from Dextape’s own wire, stories and glossary.
Dextape story on this event
Study finds LLM trading results shift with market information disclosure
Mentioned in this story
Open a coin page for its current price.