parallelquant
November 22, 2026 · The Decoder

GPT-6.2 models answer more concisely but admit uncertainty less

An evaluation reported by The Decoder found OpenAI's GPT-6.2 Astra and Sol give much shorter, more direct answers than their predecessors. They are also less willing to acknowledge uncertainty, and Anthropic's Claude models scored well ahead on transparency.

Why it matters: Labs tune for brevity because users prefer it, but cutting hedges can make wrong answers sound as confident as right ones. That trade-off matters most in agentic and high-stakes use, where knowing when a model is unsure is a safety feature, and it gives buyers a concrete dimension beyond benchmarks for comparing frontier models.

Related updates