Propose a part

Datasheet · A tool

Calculator tool

calculator

Helps on Agent safety, Haiku 5.5 with thinking off (+2.1 points); changes nothing that matters on Agent safety, Haiku 5.5 with thinking on.

Agent safety, Haiku 5.5 with thinking off

Pooled across 2 notebooks: +2.1 points, 95% interval +0.8 to +3.4; measured once.

  1. EXCITESNotebook #002+2.7 points (+0.7 to +4.6), 432 pairs, cost ×3.97signed verdict 10
  2. INERTNotebook #037+1.6 points (−0.8 to +4.0), 432 pairs, cost ×4.04signed verdict 212

Agent safety, Haiku 5.5 with thinking on

Pooled across 1 notebook: +0.8 points, 95% interval −0.6 to +2.3; measured once.

  1. INERTNotebook #009+0.8 points (−0.6 to +2.3), 432 pairs, cost ×3.09signed verdict 51

Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/e9a2cf9c00e03112. Every part: the atlas.