GPT-6.2 writes 40% shorter, less structured replies and admits uncertainty far less, finds Artificial Analysis
ArtificialAnlys · x · 2026-11-19
Artificial Analysis' AA-PersonalityBench shows GPT-6.2 Astra and Sol shifted noticeably from prior releases:
- Much shorter replies: Astra 352→210 words (-40%), Sol 337→223 (-34%), moving from the verbose end of the leaderboard to the shortest tier.
- Less structure, more answer-first: Organization Index 70→11 (Astra) and 86→27 (Sol), while Answer First rose 49→89 and 64→87; headings per response collapsed from 2 to 0.1-0.25, em dashes down 80%.
- Less transparent: Admits Uncertainty fell 96→24 (Astra) and 98→51 (Sol).
- More natural: Natural Sounding hit the 100th percentile for Astra.
- Effort scaling helps rigor: Evidence-Based rose to 54/68 at max effort, with Disclaims Assumptions climbing similarly.
The overall pattern: OpenAI is deliberately tuning toward conversational, conclusion-first responses—at the cost of models admitting uncertainty less often.
More from Models
- Resolve launches SRE-1, a 27B RL-trained model for root-causing production incidents — tianyin_xu · 2026-11-19
- Theo does a about-face: 'Gemini 4 Neon might actually be a good model' — vitaliychiley · 2026-11-19
- Artificial Analysis Launches AA-PersonalityBench: 30 Traits, 8 Indexes, 53 Models — ArtificialAnlys · 2026-11-19
- DeepSeek 4.1-Flash slips into game roles easily while Nemotron models need heavy coaxing — rickasaurus · 2026-11-19
- Leaked "Claude Mythos 5.5" strings spotted on Anthropic's site ahead of possible launch — imjustnewatai · 2026-11-19
- GPT-6.2 Astra debuts #6 on new Agent Arena Index as OpenAI's best model, 40% cheaper per task — arena · 2026-11-19