OpenAI describes the update as evolutionary rather than revolutionary. The deployment of ChatGPT 5.2 has kicked off and will appear first for subscribers on paid tiers, including Plus, Pro, Go, Business, and Enterprise plans. Free users, however, face a wait, with no immediate timeline disclosed for broader access.

OpenAI has rolled out ChatGPT 5.2. The update brings significant upgrade for professional tasks and long-running automated agents. The new variant brings reduced hallucinations and enhanced reasoning capabilities in coding, data analysis, and creative projects. It is an upgrade to ChatGPT 5.1 that was released last month.

OpenAI describes the update as evolutionary rather than revolutionary. The deployment of ChatGPT 5.2 has kicked off, and will appear first for subscribers on paid tiers, including Plus, Pro, Go, Business, and Enterprise plans. Free users, however, face a wait, with no immediate timeline disclosed for broader access.

"We're rolling this out gradually to paid plans to maintain reliability," an OpenAI spokesperson stated in the announcement. On the developer front, the API versions - labelled gpt-5.2, gpt-5.2-chat-latest, and gpt-5.2-pro—are available straight away to all registered users.

ChatGPT 5.2 pricing

GPT‑5.2 is priced at $1.75/1M input tokens and $14/1M output tokens, with a 90 percent discount on cached inputs. On multiple agentic evals, we found that despite GPT‑5.2’s greater cost per token, the cost of attaining a given level of quality ended up less expensive due to GPT‑5.2’s greater token efficiency.

OpenAI says that GPT‑5.2 is priced higher per token than GPT‑5.1 because it is a more capable model. It’s still priced below other frontier models, the company argues. If you are not willing to pay more, stick to ChatGPT 5.1

ChatGPT 5.2 vs ChatGPT 5.1: What's The Difference?

ChatGPT 5.2 refines the core strengths of its predecessor while addressing key weaknesses, particularly in reliability and multimodal processing. Where GPT-5.1 occasionally faltered on long-context retention or visual interpretation, the new model claims a 30 percent reduction in hallucinations—manifesting as factual errors—across de-identified queries involving search and maximum reasoning efforts.

Reasoning depth has been elevated, with a new "higher" effort level that excels at multi-step problems, natural idea connections, and clear breakdowns of intricate topics. Real-time data integration feels smoother, allowing the AI to track evolving discussions on subjects like live events or financial markets without losing thread. Backend tweaks also yield faster responses, especially in extended conversations, making it more practical for iterative tasks.

Multimodal capabilities see a notable boost: beyond basic image handling in GPT-5.1, 5.2 tackles enhanced formats such as charts, diagrams, and handwritten notes with halved error rates in areas like graph reasoning and interface navigation. Memory functions are now adaptive and predictive, quicker to learn user preferences and more steadfast in applying them—shifting from "custom and retentive" to a more intuitive system.

Tool access has expanded, facilitating seamless transitions between coding, design, writing, and data manipulation, which OpenAI dubs "improved workflows." For professionals, this means fewer handoffs in agentic setups, where the AI autonomously manages tools over prolonged sessions.

ChatGPT 5.2 vs ChatGPT 5.1: Benchmark results

According to OpenAI internal testing, GDPval benchmark, which tests occupational tasks across 44 professions, GPT-5.2 secured 70.9 percent wins or ties against rivals, up from 38.8 percent for GPT-5.1. Coding prowess improved on SWE-Bench Pro (55.6 percent versus 50.8 percent), while factual accuracy on GPQA Diamond rose to 92.4 percent from 88.1 percent.

In mathematics, it achieved a perfect 100 percent on AIME 2025, compared to 94 percent previously, and long-context recall neared 100 percent on the 4-needle MRCRv2 test at 256,000 tokens. Vision tasks halved errors on CharXiv and ScreenSpot-Pro, and tool-calling accuracy hit 98.7 percent on Tau2-bench Telecom, surpassing 95.6 percent.

Internal tests on spreadsheet tasks for investment banking showed a 9.3 percentage point jump to 68.4 percent completion rates.