GPT-5.2 is Here

OpenAI has launched GPT-5.2, positioning it explicitly as a model for professional work rather than general-purpose use. The company’s benchmarks show notable improvements: 55.6% on SWE-Bench Pro (up from Opus 4.5’s 52%), 52.9% on ARC-AGI2 (compared to Opus 4.5’s 37.6%), and a headline score of 70.9% on GDP-Val—OpenAI’s internal measure of economically valuable knowledge work tasks, up dramatically from 38.8% with GPT-5. CEO of Applications Fiji Simo, Greg Brockman, and head of ChatGPT Nick Turley all emphasized professional productivity: building spreadsheets and presentations, reviewing production code, analyzing long documents, and executing complex multi-step projects.

Early third-party evaluations offer a more nuanced picture. Every’s testing—which uses roughly 50 requests scored on reader engagement and avoidance of AI-isms—found that GPT-5.2 excels at instruction following and extended tasks but is less likely to surprise users, calling it a “polished professional” compared to GPT-5.1’s “brilliant, slightly chaotic freelancer.” Simon Wilson verified real improvements in structured business outputs like multi-sheet Excel workbooks and client-ready PowerPoint decks, calling this the first time ChatGPT has produced spreadsheets he would consider remotely client-ready. Ali Miller noted the model wrote code mid-task to improve its own OCR performance.

The consensus across early testers: a meaningful upgrade for enterprise and professional use cases, with some tradeoffs in creative writing voice and a tendency toward over-formatted outputs. The episode provides comprehensive coverage of both the official launch narrative and the first wave of independent community testing.

📺 Source: The AI Daily Brief: Artificial Intelligence News · Published December 12, 2025
🏷️ Format: News Analysis