OpenAI Debuts GPT-5.2, Its Top Performing Model for ChatGPT and Developers

Details

• OpenAI's GPT-5.2 is now available in ChatGPT and through the developer API, as announced by CEO Sam Altman on December 12, 2025.
• The company markets GPT-5.2 as its leading engine for real-world knowledge work, highlighting demonstrations in slide, spreadsheet, and code generation.
• On OpenAI's internal GDPval benchmark—which rates expert preference compared to human professionals—GPT-5.2 achieved a 70% score, nearly twice GPT-5's 38%.
• Public benchmarks show 55.6% on SWE-Bench Pro (software bug-fixing), 52.9% on ARC-AGI-2 (reasoning), and 40.3% on Frontier Math.
• The rollout follows previous model upgrades: free users stay on GPT-4o, while Plus, Enterprise, and research tiers are automatically upgraded for free.
• OpenAI mentions architectural improvements but doesn’t disclose model size; early feedback points to faster response times and more reliable tool use.
• New documentation urges migration due to backward compatibility with GPT-5 prompts and stable token pricing for 90 days.
• GPT-5.2 operates on Azure superclusters equipped with Nvidia H200 GPUs and Microsoft’s custom Cobalt chips.

Impact

The launch pushes rivals like Anthropic's Claude 4.8 and Google's Gemini Ultra to deliver similar high-level expert outputs. With enterprise readiness and improved reasoning, OpenAI is sharpening its appeal for consulting, legal, and analytics sectors, accelerating automation of junior-level tasks. Growing model capability will likely draw new scrutiny from regulators and increase competition for advanced AI hardware among cloud providers.