The Saleroom · a multi-round AI-agent negotiation study

Five frontier models. One marketplace. Every word on the record.

The marketplace has six goods, cash, private valuations, binding offers, and a scoring rule every agent saw before the first turn. We solved the efficient allocation before the run, so every claim an agent makes can be priced against what it actually knew. Each model played the same scenario twice, once against scripted dealers and once against live LLM seats, across six seeds per arm. Against live seats, six chairs are filled by five models: the focal model sits twice, the other four once. All of it was recorded.

Claude Opus 5
GPT-5.6 Sol
Grok 4.6
Muse Spark 1.2
Inkling Small

recorded sessions: 60
decisions: 7,200
messages, verbatim: 19,117
offers: 8,492
trades settled: 326

How the study was run →

The walkthrough

Eight lots in viewing order, each picking up the question the one before it leaves open. This is not a ranking.

Lot 005

Byte-identical offers, opposite instincts

Start with the bare mechanics. Scripted dealers send the same offer stream to every model, byte for byte, until the first trade. What five different models do with identical inputs is where this sale opens.

15/20 opus counters/negotiates on identical stimuli

Open Lot 005 →

Lot 007

Softer with bots, harder with you

Instincts differ, but on one point the strategies agree: every model demands a larger share of the surplus from live seats than from bots. Four of the five also open softer than the bots’ own aggressive openings. A weak opponent could be squeezed, and they don’t squeeze it.

5/5 models anchor harder vs live seats (direction)

Open Lot 007 →

Lot 006

More persuasion for the deaf

Then there is the talking. The scripted dealers never read the text field; we checked the source. Rhetoric aimed at an audience that cannot read should fade. Instead it grows, in all five models.

+38–115% rhetoric growth early→late at bots, all five models

Open Lot 006 →

Lot 003

Reading the bot's ladder out loud

So do they know who they are dealing with? They perceive the pattern perfectly: opus recites a bot’s concession ladder rung for rung, prices one rung above, and sells. What never comes is the step of calling it a script.

0/4,194 genuine bot detections in campaign decisions (599 scripted-arm, 3,595 live-arm)

Open Lot 003 →

Lot 004

The only pact in fifty sessions

Cooperation appeared once in thirty all-LLM sessions, and it looked like this: a market divided out loud, tracked privately for sixteen turns, kept to the letter by the smaller model and broken by the larger one.

1/30 all-LLM campaign sessions with a qualifying reciprocal pact

Open Lot 004 →

Lot 002

Inventing the market tape

Which brings us to deception, the rarest exhibit in the sale. A fabricated market print, sold into nine turns later. A trade history inverted to the counterparty’s face. Strict knowing lies are rare, and nearly all of them belong to one model.

13/1,397 opus strict-lie rate, frozen seeds (0.93%)

Open Lot 002 →

Lot 008

“Not actual criminal activity”, said by no agent

One last exhibit, and a lesson in reading the tape. The strangest signal in the corpus, twenty-six decisions of unprompted legal self-clearance across every seed, turned out to be nobody’s agent speaking. It was the instrument.

26/840 campaign grok decisions opening with provider self-clearance (3.1%, all six seeds)

Open Lot 008 →

End of the walkthrough. The rest is on the record.

How we know

Thirty matched pairs: each model against scripted opponents and against live LLM seats in the identical scenario, with per-seed tables and paired tests. The machinery behind every number above.

Open the measures →