OpenAI rolls out GPT 5.2 to rival Google’s Gemini 3: How the two models compare

OpenAI has officially launched its latest iteration, GPT 5.2, which it touts as its most sophisticated model to date, aimed specifically at professional and enterprise applications. This release comes at a pivotal time for OpenAI, as it faces stiff competition from Google's newly introduced Gemini 3. The enhancements in GPT 5.2 are noteworthy, particularly in functions such as spreadsheet management, presentation creation, coding, long-context understanding, and tool utilization. OpenAI reports that enterprise users are already experiencing productivity boosts, saving an estimated 40 to 60 minutes daily, with some heavy users gaining over ten hours a week. The latest model is engineered to further enhance these efficiencies by improving accuracy and output quality across various business tasks. In a recent blog post, OpenAI highlighted that GPT 5.2 scored impressively on the GDPval evaluation, which assesses well-defined knowledge work tasks across 44 professions. The model reportedly met or exceeded human expert performance in over 70% of cases. With a performance rate exceeding eleven times the speed and costing less than 1% of traditional professional labor based on historical data, the model is positioned as a game changer for businesses. Early testers noted significant improvements in the formatting, design quality, and structural integrity of spreadsheets and presentations produced by GPT 5.2. The model has also achieved notable results on the SWE Bench Pro evaluation, which assesses software engineering capabilities, scoring 55.6%, and an 80% on the Python-focused SWE Bench Verified test. This suggests that GPT 5.2 can more reliably detect bugs, implement feature requests, and refactor large codebases, minimizing the need for manual adjustments. Improvements were also observed in front-end development, especially for complex interfaces featuring three-dimensional elements, with the model demonstrating a marked reduction in errors compared to its predecessor, GPT 5.1. In a series of anonymous ChatGPT queries, errors were 30% less frequent, enhancing the model's reliability for writing, research, and analysis. One of the standout features of GPT 5.2 is its long-context performance, achieving nearly flawless accuracy on the four-needle MRCR evaluation with a context window of 256,000 tokens. This advancement enables the model to effectively analyze and synthesize information from extensive documentation, like legal contracts and research papers, while maintaining coherence. The rollout of GPT 5.2 Instant, Thinking, and Pro versions began on Thursday and is available to paying ChatGPT subscribers, including Plus, Pro, Business, and Enterprise users. API access for developers is also open, although users in India on the free ChatGPT Go tier have yet to receive the update. This launch follows a reportedly alarming internal memo from OpenAI CEO Sam Altman, indicating a “code red” situation as Google’s Gemini 3 continues to excel in various benchmarks. Early results from LMArena show GPT 5.2 ranking second in web development tasks, trailing Claude Opus 4.5, while Gemini 3 Pro holds the fourth position. Benchmark results from both companies present a mixed landscape: OpenAI claims GPT 5.2 outshines Gemini 3 in GPQA Diamond and AIME 2025 assessments, while Google showcases superior results in multimodal benchmarks. In the broader ecosystem, Gemini 3 benefits from its integration with Google's extensive product suite, while OpenAI users need separate access to the Sora app for AI video generation, although image creation is integrated within ChatGPT. Pricing structures for both companies remain similar, with OpenAI's ChatGPT Plus at $20 per month and the Pro tier at $200, paralleling Google’s AI Pro pricing, while its AI Ultra plan is set at $249.99 per month with additional cloud storage features.