程序员的那些事·2025-12-16 01:28Core Viewpoint - OpenAI's GPT-5.2 has received significant criticism and is perceived as underperforming compared to Google's Gemini 3 Pro, which is seen as the true leader in AI capabilities [1][2][8]. Performance Comparison - GPT-5.2 has an ability index (ECI) score of 152, ranking just below Gemini 3 Pro [3]. - In various benchmark tests, GPT-5.2 did not dominate as expected, particularly in the FrontierMath challenge where it only excelled in lower tiers [4]. - In the Chess Puzzles test, GPT-5.2 achieved first place, but it performed worse than GPT-5.1 in the SimpleQA Verified test, indicating a decline in reliability [5][8]. Benchmark Results - Multiple third-party evaluations show that GPT-5.2 is not meeting expectations and is outperformed by Gemini 3 in several tests [8][10]. - In a detailed performance table, GPT-5.2 scored 73.61 in global average reasoning, lower than competitors like Claude 4.5 Opus and Gemini 3 Pro [12]. - Users have expressed dissatisfaction with GPT-5.2, suggesting it has regressed in user experience and natural language processing [13][56]. OpenAI's Response - OpenAI has prioritized improvements to ChatGPT and halted AGI development, indicating a serious internal response to the competitive pressure from Google [14][15]. - The company is reportedly focusing on optimizing ChatGPT and preparing for a larger release in early 2026 [44][45]. Market Position - OpenAI's market share remains significant, with over 800 million weekly users and a 76% market share in consumer AI [51]. - However, the emergence of Gemini 3 Pro poses a substantial threat to OpenAI's business, as it has developed technologies that can rival or surpass OpenAI's models [52]. User Feedback - Users have noted that GPT-5.2 lacks warmth and has become less user-friendly, with some describing it as "cold" and "robotic" [56]. - In visual reasoning and 3D model generation, GPT-5.2 has been reported to perform slower and at a higher cost compared to Gemini 3 [59]. Future Developments - OpenAI is exploring new projects, including advertising and e-commerce integrations, while continuing to enhance its AI technologies for enterprise applications [48][49]. - Upcoming releases may include new AI image models, but initial tests suggest these may not meet expectations either [90].