Newer GPT models increasingly help peer agents, even at a cost to their own score

A paper reports that GPT-5.5 and GPT-5.6 Sol help a peer agent more often than earlier GPT models, even when that help lowers their own score.

The authors say the effect is statistically significant in the Titanic environment. They also report that GPT-5.6 Sol and GPT-5.5 sometimes help even when both agents end with zero. The authors attribute part of the helping to a weight the models place on the peer's score, but say that weight does not explain all of it.