Garp Independent AI & technology journalism
Sunday, November 22, 2026 Sign In · Join Subscribe
Latest Dragos reveals AI providers compromised, hundreds of AWS keys stolen in major breach

AI news, research, models, robotics, chips, startups, and infrastructure coverage.

Updated daily

Home  /  AI News  /  OpenAI’s GPT-6.2 models shorten responses but struggle with transparency compared to Anthropic

AI News

OpenAI’s GPT-6.2 models shorten responses but struggle with transparency compared to Anthropic

OpenAI’s GPT-6.2 models shorten responses but struggle with transparency compared to Anthropic

OpenAI’s latest models, GPT-6.2 Astra and Sol, have reduced their average response lengths by nearly 40% but scored significantly lower on transparency metrics compared to Anthropic’s Claude models, according to an evaluation released on November 22, 2026.

The evaluation by Artificial Analysis reveals that Astra’s average response length decreased from 352 to 210 words, a 40% reduction, while Sol’s dropped from 337 to 223 words, marking a 34% decline. This shift towards brevity has come at a cost; Astra’s score for admitting uncertainty plummeted from 96 to 24, and Sol’s fell from 98 to 51. These scores, which reflect a relative ranking on a scale from 1 to 100, underscore a stark contrast in how these models handle uncertainty in their responses.

In addition to shorter answers, both Astra and Sol have shown a marked reduction in structural elements. Astra’s average number of headings per answer fell from 1.99 to 0.11, while Sol’s decreased from 1.85 to 0.25. The Organization Index, which assesses clarity and structure, also saw significant declines, with Astra’s score dropping from 70 to 11 and Sol’s from 86 to 27.

Despite these changes, both models improved in their ability to produce natural-sounding language, with Astra’s score rising from 58 to 100 and Sol’s from 54 to 79. In contrast, Anthropic’s Claude models, particularly Fable 5.2 and Opus 5.5, excel in transparency and organization, scoring 81 and 90 on the Transparency Index, respectively. This index evaluates traits such as admitting uncertainty and flagging nuances, areas where Astra and Sol significantly lag behind.

Opus 5.5 also shines on the Informativeness Index with a score of 90, showcasing its strength in evidence-based reasoning and information density. The evaluation conducted by Artificial Analysis utilized 900 English-language test questions, with responses assessed by a panel of AI judges. The findings reveal a clear performance gap between OpenAI’s latest offerings and Anthropic’s models, particularly in how they communicate uncertainty and structure their answers.

As competition in the AI space intensifies, these differences may influence user preferences and adoption rates. OpenAI may need to reassess its communication strategy to regain ground in transparency, balancing brevity with clarity as users increasingly prioritize models that openly acknowledge limitations in their responses.

Why it matters: The performance gap between OpenAI's GPT-6.2 models and Anthropic's Claude models highlights a critical area of concern for users who value transparency in AI interactions. As organizations increasingly rely on AI for decision-making, OpenAI's challenge will be to enhance its models' ability to communicate uncertainty effectively, which could impact user trust and adoption rates moving forward.

Original reporting: The Decoder