OpenAI

Model vendor that released GeneBench-Pro (Jun 30 2026), a high-quality synthetic-ground-truth genomics reasoning benchmark. Quality_median 5 for the single tracked benchmark; openness 2 (only 10 of 129 problems public). Flagged model_vendor: it authors both a leading model family and the benchmark used to rank models, so treat cross-model comparisons with that context. Narrow bio-benchmark output to date (breadth 2).