OpenAI GPT-6 Sol and Luna System CardProvider report
2026-09-22Sixteen scores across eight capability benchmarks from the Sol/Luna appendix added to the GPT-6 system card on September 22, 2026. The four HealthBench metrics use Table 29's length-adjusted scores separately from unadjusted results; raw scores and mean response lengths are preserved as supplementary data. Cybersecurity results come from Appendix A.8.1.2. ExploitBench retains the historical-vulnerability contamination caveat, and the internal June-August 2026 set is separate, including Luna's genuine zero. Effort is recorded only when stated. Comparison models and unlabelled curve estimates are excluded.