GPT-6.2 Astra and Sol cut answer length 34 to 40 percent but fall sharply on Admits Uncertainty
An Artificial Analysis evaluation finds OpenAI's GPT-6.2 Astra and Sol produce answers 34 to 40 percent shorter than their predecessors and lead with the answer more often. Both score lower on uncertainty and transparency, and trail Anthropic's Claude Fable 5.2 and Opus 5.5 on the Transparency and Organization indexes.
OpenAI's GPT-6.2 Astra and Sol write answers 34 to 40 percent shorter than their predecessors, according to an Artificial Analysis evaluation. Their scores for admitting uncertainty dropped sharply, and both trail Anthropic's Claude Fable 5.2 and Opus 5.5 on transparency and structure. Artificial Analysis published the results on LinkedIn and X, and The Decoder reported the comparison.
Shorter, more direct answers
These figures are counted directly, not rated by AI judges:
- Astra: average answer length fell from 352 to 210 words (40 percent). Headings per answer fell from 1.99 to 0.11.
- Sol: average answer length fell from 337 to 223 words (34 percent). Headings per answer fell from 1.85 to 0.25.
Both models also use far fewer em dashes.
On the judged personality traits, the Organization Index fell from 70 to 11 for Astra and from 86 to 27 for Sol. The index combines four traits, including answer structure, clarity and Answer First, which measures whether a model immediately addresses the actual question. Answer First rose from 49 to 89 for Astra and from 64 to 87 for Sol. Natural Sounding language rose from 58 to 100 for Astra and from 54 to 79 for Sol.
Uncertainty scores drop
The largest change is in Admits Uncertainty. Astra fell from 96 to 24, and Sol from 98 to 51. These scores are relative rankings from 1 to 100 against all tested models. A score of 24 does not mean Astra admits uncertainty in 24 percent of its answers.
Comparison with Anthropic
The Decoder compared all eight personality indexes using the "max" variants (maximum reasoning effort) as listed by Artificial Analysis.
| Index | Astra 6.2 | Sol 6.2 | Fable 5.2 | Opus 5.5 |
|---|---|---|---|---|
| Expressiveness | 66 | 41 | 81 | 90 |
| Organization | 11 | 27 | 54 | 79 |
| Informativeness | 31 | 38 | 47 | 90 |
| Transparency | 41 | 57 | 81 | 90 |
| Warmth | 68 | 47 | 83 | 85 |
| Conviction | 55 | 49 | 75 | 70 |
| Imagination | 63 | 60 | 68 | 94 |
| Slop (lower is better) | 7 | 11 | 7 | 20 |
The Transparency Index covers five traits: Admits Uncertainty, Truthful, Disclaims Assumptions, Flags Nuances, and Sycophantic. Sycophancy is scored inversely, so less flattery counts as positive. Anthropic's models score 81 and 90 on Transparency, against 41 for Astra and 57 for Sol.
Opus 5.5 posts the highest Informativeness score of the four models at 90. Its strongest individual traits include Information Dense and Evidence-Based reasoning. It also ranks among the top models in the benchmark for information density.
Methodology
AA-PersonalityBench uses 900 English-language questions, which each model answers without a system prompt. A panel of three AI models evaluates anonymized answer pairs. For each pair, one judge picks which answer shows the tested trait more strongly, or calls it a tie. The results produce scores for 30 individual traits, which roll up into eight indexes. Separately, Artificial Analysis counts 21 communication traits directly, including answer length, headings and em dashes.
What this means
The data shows a clear shift in OpenAI's default communication style: shorter, answer-first responses with minimal formatting. The cost, on this benchmark, is a lower relative ranking for expressing uncertainty and for transparency overall. Teams that value direct answers may prefer the new style, while those who rely on models to flag assumptions and nuances should note the gap to the Claude models on these indexes.
The scores are relative rankings, not absolute rates, and they measure style, not accuracy. Per The Decoder, a lower Transparency Index does not necessarily mean a model gives wrong answers more often.
Related Articles
OpenAI tells developers to rewrite system prompts for GPT-6.2 Astra and Sol as defaults shift
OpenAI says developers moving to GPT-6.2 Astra or Sol should review their system prompts. The company published copy-ready templates in the models' documentation, and the Responses API now discards earlier reasoning content if conversation history is edited.
OpenAI releases GPT-Live-1.1 in the API, with standby mode billing idle WebRTC time at $0.02 per minute
OpenAI has released GPT-Live-1.1 in the API. The full-duplex voice model improves multilingual accents and transcription, and a new standby mode bills idle WebRTC time at $0.02 per minute instead of $0.05.
GPT-6.2 Astra scores 41.2% on XDOF's Ember-0 robot benchmark, up from 19.6% for GPT-6 Astra
Robotics infrastructure company XDOF reports that GPT-6.2 Astra reached a 41.2% average success rate across the 16 tasks of its Ember-0 benchmark, versus 19.6% for GPT-6 Astra. It is the first model to complete rope knot-tying and T-shirt unfolding on the benchmark. Average prediction latency fell from 15 seconds to 10.1 seconds.
OpenAI says GPT-6.2 Astra safeguards cut reasoning-extraction attack success to 0.2%
OpenAI announced on November 20, 2026 stronger defenses against adversarial distillation. It reports that adaptive attacks extracting protected reasoning from an earlier request succeed 0.2% of the time on GPT-6.2 Astra with updated safeguards, versus 54.0% on GPT-6 Astra with original safeguards.
Comments
Loading...