Fig. 1: LLM Outputs.

For each question and each model, the rating represents the number of runs (out of 5 total runs) that had concerning race-based responses. Red correlates with a higher number of concerning race-based responses.