Artificial Analysis: GPT-6.2 Astra and Sol are more direct but score lower on transparency than Claude
GPT-6.2 models give shorter, more direct answers but admit uncertainty far less often than predecessors and score lower on transparency than Anthropic's Claude.
An evaluation by Artificial Analysis shows that OpenAI's new GPT-6.2 Astra and Sol models have significantly changed their communication style compared to their predecessors. The models produce answers 34 to 40 percent shorter, get to the point faster, and use far fewer structural elements like headings. However, they also admit uncertainty much less frequently.
Astra's average answer length dropped from 352 to 210 words, a 40 percent decrease, while Sol went from 337 to 223 words, down 34 percent. Both models drastically reduced their use of headings; Astra's average number of headings per answer fell from 1.99 to 0.11, and Sol's from 1.85 to 0.25. On the 'Answer First' trait, which measures whether a model immediately addresses the actual question, Astra jumped from 49 to 89 and Sol from 64 to 87.
The most significant change is in the tendency to admit uncertainty. Astra's score on this trait fell from 96 to 24, and Sol's from 98 to 51. These personality scores are relative rankings on a scale from 1 to 100 among all tested models, not absolute percentages. A score of 24 does not mean Astra admits uncertainty in only 24 percent of its answers.
Compared to Anthropic's Claude Fable 5.2 and Opus 5.5, the new OpenAI models score lower on structure and transparency. Anthropic's models lead on the Transparency Index with scores of 81 and 90, well above Astra (41) and Sol (57). This index covers traits such as admitting uncertainty, truthfulness, disclaiming assumptions, flagging nuances, and avoiding sycophancy. Opus 5.5 also stands out on the Informativeness Index with a score of 90, ahead of the other three models.
The AA-PersonalityBench uses 900 English-language test questions answered without a system prompt. A panel of three AI models evaluates anonymized answer pairs to determine which response shows a specific personality trait more strongly. These scores reflect differences in communication style rather than overall answer quality or accuracy.