OpenAI has officially launched GPT-5.2, its most powerful frontier AI model to date, and it is now beginning to roll out to ChatGPT users. Following what was reportedly treated as a “code red” moment internally, GPT-5.2 is being positioned as OpenAI’s direct response to Google’s Gemini 3 Pro. The update introduces three distinct model variants: GPT-5.2 Instant, GPT-5.2 Thinking, and GPT-5.2 Pro.
What’s new in ChatGPT 5.2?
According to OpenAI, the GPT-5.2 lineup is designed for advanced professional use cases and long-running autonomous agents. For knowledge-intensive tasks, GPT-5.2 delivers OpenAI’s strongest performance yet. The model excels at writing and debugging code, generating spreadsheets, creating presentations, processing long-context information, analyzing visual inputs, and efficiently using built-in tools.
GPT-5.2 is also capable of managing complex, multi-step workflows while switching between tools seamlessly. In benchmark testing, the model reportedly outperforms competitors such as Google’s Gemini 3 Pro and Anthropic’s Claude Opus 4.5 across a wide range of evaluations, reinforcing its position at the top of the current AI model landscape.
In the SWE-Bench Pro benchmark, which evaluates performance on real-world software engineering tasks, GPT-5.2 Thinking recorded a score of 55.6 percent. This places it ahead of Claude Opus 4.5, which scored 52.0 percent, and Google’s Gemini 3 Pro at 43.3 percent. Moving to SWE-Bench Verified, GPT-5.2 Thinking achieved an 80.0 percent score, though Claude Opus 4.5 narrowly retained the top position with 80.9 percent.
On the GPQA Diamond benchmark, which focuses on advanced science-based questions, GPT-5.2 Thinking reached 92.4 percent without relying on external tools, once again outperforming Gemini 3 Pro. The model also delivered strong results on ARC-AGI-2, scoring 52.9 percent—well above Claude Opus 4.5 at 37.6 percent and Gemini 3 Pro at 31.1 percent.
On the GDPval benchmark, which evaluates structured knowledge work across 44 different professions, GPT-5.2 Thinking matches or surpasses expert-level human performance on 70.9 percent of evaluated tasks.
ChatGPT 5.2 is rolling out today for paid subscribers across Go, Plus, Pro, Business, and Enterprise plans. The model’s knowledge cutoff is August 2025, making it one of OpenAI’s most up-to-date AI systems so far. Additionally, OpenAI has indicated that an adult mode for ChatGPT is planned for release in the first quarter of 2026. The company is also reportedly testing an age-estimation system aimed at identifying users under the age of 18.
