
What happened
Artificial Analysis's evaluation found OpenAI's GPT-6.2 Astra and Sol give shorter, more direct answers than predecessors, but the Admits Uncertainty trait fell from 96 to 24 for Astra and 98 to 51 for Sol.
Why it matters
The trade-off is clear: faster, more direct responses coincide with a much lower relative tendency to admit uncertainty, which could shift how much caveat readers should expect.
WHO IT HITSBusiness readers who rely on AI assistants for research or customer-facing summaries may need to recalibrate how much hedging they expect from OpenAI models. Teams comparing assistants for procurement should note that style scores describe communication, not answer accuracy.
Summaries like this, every morning.
The evaluation details a broader shift in OpenAI's communication style. Beyond the drop in Admits Uncertainty, Astra's average number of headings per answer fell from 1.99 to 0.11, and Sol's from 1.85 to 0.25. The Organization Index also dropped sharply—Astra from 70 to 11 and Sol from 86 to 27—while Answer First, a measure of whether a model immediately addresses the actual question, jumped from 49 to 89 for Astra and 64 to 87 for Sol. Scores for Natural Sounding language also climbed.
Artificial Analysis used 900 English-language test questions answered without a system prompt. A panel of three AI models evaluated anonymized answer pairs to rate 30 individual personality traits. Those traits feed into eight higher-level indexes, and the index values from 1 to 100 reflect each model's relative position among all tested models.
The scores describe communication style, not overall answer quality. A lower transparency score does not necessarily mean a model gives wrong answers more often.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
FlashLabs updated Hermes Agent, its AI agent inside the FlashRev B2B sales platform, adding multiple sessions…

Between 1:48 and 5:21 PT on November 21, requests to OpenAI's API and ChatGPT failed at rates peaking at 75%…

OrcaRouter started providing an API for Qwen4 Max, Alibaba's latest flagship model, so developers can call it…

Education Minister Choi Kyo-jin resigned Sunday after the top scorer in a national tax accountant exam confess…

A developer's task.md grew to 1,630 lines, about 45% records of finished work, and Claude reported 63 unfinish…

A Zenn author published a master index of every article in their series — Rokid Glasses smart glasses (15 post…
