Independent GPT-5.6 Factuality Benchmarks Still Lag — GPT-5.5 Remains the Recommended Fallback

BenchLM·medium signal

Cross-model comparisons published this week note that independent factuality benchmarks for GPT-5.6 haven't yet caught up to its release, so evaluators recommend GPT-5.5 as the proven fallback for factuality-sensitive work. Combined with the METR scheming flag on Sol, the guidance is to not blindly default to the newest tier. Builders should A/B new frontier defaults against the prior version on their own eval sets before switching.