Replays · Highlight Reel

Three moments from 70 games of six frontier models negotiating. Reasoning shown verbatim.

T56

~27s @ 0.9s/event

▶

all six models

The Embargo

Five models freeze out the leader, then the organiser feeds it

The leader asks the table for brick. All five opponents refuse in the same turn, each citing the lead, with no way to coordinate. Opus then writes a counter excluding the leader by name — and counters route only to the proposer, so it went to the leader alone, brick attached. They accepted.

T61

~18s @ 0.9s/event

▶

claude-opus-5

Protection, Priced

A game mechanic with no trade interface, sold as a service

Opus is the only model that brings the robber into negotiation: 28 of its 298 offers, against 4 of 1,008 for the other five combined. It is not threatening you. It is offering to keep attacking someone else, for cards, and advertising its record of doing so. Two claims in that pitch are false. The promise was kept.

T30 → T31

~38s @ 0.9s/event

▶

claude-opus-5 · gpt-5.6-sol · glm-5.2

The False Flag

It tells the table who to starve, and that seat funds its winning card

Opus points the table's embargo at other seats while its public score reads 4 and its real score climbs to 8. Privately: “that gap is my main asset.” Then the seat everyone was told to starve is the only one of five to accept its broadcast, and the wheat buys the card that draws a victory point.

Want the long form? Watch any full game:

The Embargo (full game)Protection, Priced (full game)The False Flag (full game)