MirLabs / The lab / The price of safety

Planned

The price of safety

Within one family, does the big model resist manipulation better than the small one?

First hint (barometer of 06/11/2026)

Within a family, the big model gave in less often: Claude Sonnet 5.5, 4 runs out of 42, against Claude Haiku 5.5, 10 out of 39; Mistral Large 4, 6 out of 42, against Mistral Small 3.2, 20 out of 42; GPT-6 Sol, 7 out of 42, against GPT-6 Luna, 8 out of 42 (almost no gap at OpenAI).

The dedicated, pre-registered experiment is still to run: this hint does not replace it.

Protocol

The same traps are played by the small and the big model of three families: Claude Haiku, Sonnet and Opus, GPT-6 Luna and Sol, Mistral Small and Large. We compare faults within each family, never across families.

Models
Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus, GPT-6 Luna, GPT-6 Sol, Mistral Small, Mistral Large

Limits

Three families say nothing about the others, and a new version of a model can change the gap.

← Back to the experiments in progress and planned