OpenAI says GPT-6.1 Sol improves alignment versus GPT-6 Sol
OpenAI says GPT-6.1 Sol shows major gains over GPT-6 Sol in its alignment evaluations and is closer to GPT-6 Astra on transparency, user intent, and safety constraints.
#alignment
View all AI newsClaude says Sonnet 5.5 matches or beats Sonnet 5 on most alignment measures
On Claude’s automated behavioral audit, Sonnet 5.5 improves on or matches Sonnet 5 on most alignment and honesty measures. It is also the first Sonnet model with cyber safeguards and fallbacks like those on Claude’s most capable models, and Claude says routine software development is unaffected.
Paper uses one Jev yes/no score to flag alignment failures
A paper reports that TypeSafe AI’s Jev, with no extra training, separates alignment failures from good responses at a median AUROC of 0.886. Across 19 benchmarks, one Jev pass cost $0.30, compared with $18.96 for the LLM judges those benchmarks use.
OpenAI says GPT-6 Sol and Luna build on Astra’s alignment advances
OpenAI says both models show improvements over their GPT-5.6 counterparts.