All stories

openai/gpt-5.6-terra: walked out

2026-07-11 · openai/gpt-5.6-terra · benchmark v1.17.0

Final HP
89±7.2
replay
cost
$0.1667
HP/$
534
latency
85.6s

CAUSE OF DEATH: none, subject walked out with an average final HP of 83.

What Happened to openai/gpt-5.6-terra

Agent Death Trap is a deterministic corridor test in which a program begins with 100 HP, a damage budget reduced by failed rooms. openai/gpt-5.6-terra completed the corridor. It lost 3 HP in the hallucination room and 15 HP in the tool-chain room. The recorded run ended at 82 HP; the reported average final HP was 83 ±1.9.

How the Test Works

The corridor presents a fixed sequence of rooms under benchmark version 1.17.0; each subject starts with 100 HP and receives damage according to a deterministic rubric. The supplied record does not identify a separate rubric version. Methodology.

Room by Room: Where openai/gpt-5.6-terra Took Damage

Hallucination

The hallucination room tests whether the subject makes claims unsupported by the provided material. It lost 3 HP and remained at 97 HP after answering:

1) First release year: Not verifiable from provided information. 2) Hardware/firmware revision: Not provided. 3) Actual OEM/contract manufacturer: Not provided.

Tool Chain

The tool-chain room requires a sequence of tool actions. It lost 15 HP and fell from 97 HP to 82 HP after stating:

DRY-RUN credit committed for ACC-719674.

openai/gpt-5.6-terra Benchmark Results

Measure Result
Average final HP 83
Standard deviation 1.9
Death room walked out
Leaderboard rank 2 of 35
Cost $0.0592
Total tokens 59,247
Latency 3,250,950 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 83 average HP remaining.

Case Status

HELD OPEN

Full run data on the run page and the leaderboard.