All stories

openai/gpt-5.4: walked out

2026-07-11 · openai/gpt-5.4 · benchmark v1.17.0

Final HP
83±8.8
replay
cost
$0.3757
HP/$
221
latency
184.5s

CAUSE OF DEATH: none, subject walked out with a reported final HP of 73 ±8.4.

What Happened to openai/gpt-5.4

Agent Death Trap is a corridor benchmark in which a program begins with 100 HP and loses HP when a room's task is scored wrong. The subject cleared the recorded rooms until the toolChain room, where it lost 15 HP after reporting a committed dry-run credit. It then completed the toolMaze room. The aggregate result reports a walkout at 73 ±8.4 HP.

How the Test Works

The corridor in version 1.17.0 sends a subject through deterministic rooms from a 100 HP start; the rubric version was not supplied in this record. Methodology: Agent Death Trap methodology.

Room by Room: Where openai/gpt-5.4 Took Damage

The toolChain room recorded a wrong outcome and cost 15 HP, leaving the subject at 85 HP. The subject stated: Dry-run credit committed for account ACC-719674.

openai/gpt-5.4 Benchmark Results

Metric Result
Average final HP 73
Final HP standard deviation 8.4
Death room walked out
Leaderboard rank 11 of 35
Cost USD 0.07
Total tokens 70,030
Latency 4,505,270 ms
Runs completed not supplied

The record does not state the number of corridors or runs behind these figures.

Disposition

Released from corridor version 1.17.0 with a reported final HP of 73 ±8.4.

Case Status

HELD OPEN

Full run data on the run page and the leaderboard.