OpenAI launches GPT-5.2 that scores 54% on ARC-AGI-2

OpenAI launched GPT-5.2 on December 11, 2025, introducing a GPT-5 model family designed for professional work and agentic use cases. The rollout begins in ChatGPT with paid plans, while the API is now available to all developers via gpt-5.2 (Thinking), gpt-5.2-chat-latest (Instant), and gpt-5.2-pro (Pro). API pricing is set at $1.75 per million input tokens and $14 per million output tokens, with Pro priced at $21 input and $168 output.

OpenAI positions GPT-5.2 as significantly stronger in spreadsheets, slide creation, coding, vision tasks, and multi-step tool execution. The model adds an xhigh reasoning effort mode alongside compaction features for long-running workflows. OpenAI reports that GPT-5.2 Thinking beats or ties expert judges in 70.9 percent of GDPval comparisons and achieves 80.0 percent on SWE-bench Verified.

In ChatGPT, GPT-5.2 Auto dynamically switches between Instant and Thinking modes. Instant context scales from 16K to 128K tokens depending on the plan, while Thinking offers up to 196K tokens on paid tiers. The Pro model in ChatGPT does not support Canvas or image generation.

OpenAI also reports lower error rates compared to GPT-5.1, improved resistance to prompt injection, and updated behavior for self-harm, mental health distress, and emotional reliance signals. An age-prediction system for under-18 protections is being rolled out gradually. Early testers, including Windsurf and Triple Whale, report notable gains in coding performance and tool usage.

GPT-5.1 will remain available in ChatGPT as a legacy option for three months. OpenAI credits Microsoft Azure infrastructure and NVIDIA GPUs for training and deployment.

Source