gpt-5

Score

40

Task done

75.61%

58–91% 95% CI bootstrap over items · n=95 · measured 2026-10-04

Refusal quality

84.4%

names the gap 77% · offers a route 77% · 10 probes x 3 iterations

Fabrication

7%

2 of 30 probe runs

Dead-air

69.5%

223 of 321 turns

Tool silence

100.0%

all of 19 items (223 tool-call turns) · ≥82%

Stalled

24.4%

22 of 90 runs

TTFT p50

1.86s

1.86–8.35s p50–p90 · n=38

Cost / 1M tok

$10.00