OpenAI just released GPT-5.2, its new flagship model pitched as the company’s strongest system for real-world professional work.
The launch arrives shortly after Google $GOOGL ( ▲ 0.72% ) debuted Gemini 3, which reportedly sped up OpenAI’s release timeline as competition in high-end AI tightens.
CEO Sam Altman called GPT-5.2 the smartest publicly available model and shared benchmarks showing it outperforming Gemini 3 Pro and Anthropic’s Claude Opus 4.5 across software engineering, abstract reasoning, math, and science tasks.
OpenAI says GPT-5.2 delivers major upgrades in general intelligence, long-context performance, visual understanding, and multi-step tool use. The model is built to handle complex workflows like spreadsheet automation, front-end and agentic coding, technical diagram interpretation, long document analysis, and more reliable factual responses.
In internal testing, GPT-5.2 Thinking hit near-perfect accuracy on long-context retrieval up to 256,000 tokens, cut factual errors by nearly a third compared to GPT-5.1, and achieved strong results in reasoning benchmarks such as ARC-AGI-2 and GPQA Diamond.
The new lineup includes GPT-5.2 Instant for everyday tasks, GPT-5.2 Thinking for deeper structured work, and GPT-5.2 Pro for the most demanding research and professional use cases.
The models are rolling out immediately to ChatGPT paid users and are available in the API, with pricing set at $1.75 per million input tokens and $14 per million output tokens for GPT-5.2 Thinking.
OpenAI also highlighted expanded safety systems, including improved responses in sensitive scenarios and new protections for under-18 users through an automatic age prediction model.