OpenAI’s GPT-6.2 Astra and Sol have a more concise communication style than their predecessors, according to an evaluation by Artificial Analysis. Their answers are shorter and more direct, but the models scored lower on admitting uncertainty. A comparison with Anthropic’s Claude models also found differences in transparency, organization, and informativeness.
Shorter answers, fewer headings
Astra’s average answer length fell from 352 to 210 words, a 40 percent decrease. Sol’s dropped from 337 to 223 words, down 34 percent. The evaluation also counted fewer headings in responses: Astra went from an average of 1.99 to 0.11 per answer, while Sol went from 1.85 to 0.25.
Those counts suggest a visible change in presentation. Both newer models use less formatting to break up their replies, and their answers take fewer words on average. The figures describe response style; on their own, they do not show whether a shorter answer is more useful for a particular question.
More direct, but less likely to signal uncertainty
Some of the style changes came alongside higher scores for addressing questions directly. Astra’s Answer First score rose from 49 to 89, and Sol’s from 64 to 87. This measure looks at whether a model immediately addresses the question. Natural Sounding language scores also increased: from 58 to 100 for Astra and from 54 to 79 for Sol.
At the same time, the models’ scores for admitting uncertainty fell sharply. Astra dropped from 96 to 24, while Sol fell from 98 to 51. These are relative rankings on a scale from 1 to 100, not percentages of answers in which a model admits uncertainty. A score of 24, for example, does not mean Astra expresses uncertainty in 24 percent of its answers.
The Organization Index also declined, from 70 to 11 for Astra and from 86 to 27 for Sol. Artificial Analysis says the index combines four traits rated by AI judges, including answer structure, clarity, and Answer First. That combination helps explain why a model can become more direct while still receiving a lower overall organization score.
How the Claude models compare
In the comparison of the max model variants listed by Artificial Analysis, Claude Fable 5.2 and Opus 5.5 scored higher on the Transparency Index than Astra and Sol. The scores were 81 and 90 for Anthropic’s models, compared with 41 for Astra and 57 for Sol.
The Transparency Index brings together five traits: Admits Uncertainty, Truthful, Disclaims Assumptions, Flags Nuances, and Sycophantic. The sycophancy trait is scored inversely, so less flattery counts as a positive. Anthropic’s models also led on Organization in the comparison. Opus 5.5 stood out on the Informativeness Index, scoring 90; its strongest individual traits included Information Dense and Evidence-Based reasoning.
What the evaluation can—and cannot—tell readers
Artificial Analysis’s AA-PersonalityBench uses 900 English-language test questions, which each model answers without a system prompt. Three AI models evaluate anonymized answer pairs. In each comparison, one judge selects which answer displays the tested personality trait more strongly, or calls it a tie.
The evaluations produce scores for 30 personality traits, which feed into eight higher-level indexes. The index values reflect each model’s relative position among all tested models. Separately, the benchmark directly counts 21 communication traits, including average answer length and the number of headings and em dashes.
These results describe communication tendencies, not overall answer quality. A lower Transparency Index does not, by itself, mean a model gives wrong answers more often. The findings instead highlight a trade-off in how the models present information: Astra and Sol became more concise and direct, while scoring lower on measures related to uncertainty and transparency.