We are closely tracking the rapid evolution of generative AI capabilities, and the introduction of the GPT-5.2 model marks a significant advancement in this landscape. As competitive pressure heightens across the sector, OpenAI has released its latest model with claims of substantial improvements in professional reasoning, coding performance, and advanced task execution. The arrival of the GPT-5.2 model comes less than a month after the debut of GPT-5.1, highlighting the accelerating pace of AI innovation.

The new version demonstrates strong performance across benchmarks covering complex knowledge work, including legal analysis, financial modeling, accounting, mathematical reasoning, and multi-step coding tasks. According to data shared by OpenAI, results surpass earlier iterations across a range of accuracy and reliability metrics. This development arrives as key competitors such as Google with Gemini 3 Pro and Anthropic with Claude Opus 4.5, continue expanding enterprise adoption.

Fidji Simo, CEO of Applications at OpenAI, clarified that this release is not positioned as a direct reaction to Gemini 3 Pro. Instead, she emphasized that the new system had been in development for many months, underscoring the long-term planning involved in building large foundation models. Internally, the model carried the code name “Garlic,” referenced publicly after OpenAI CEO Sam Altman posted a teaser video featuring the ingredient the day before launch.

Development Continued Despite Internal Priority Shifts

Simo explained that the organization’s earlier “code red” declaration initiated in response to competitive dynamics, did not determine the release timing. The new system had already reached testing phases before the internal directive was announced. The “code red” instruction instead focused teams on strengthening ChatGPT and refining core product experiences.

Early evaluations were carried out by “Alpha customers,” including Harvey, Notion, Box, Shopify, and Zoom. These organizations assessed the model across real-world workflows for several weeks. Feedback highlighted improvements in tool use, coding support, task planning, and structured output generation.

Coding remains one of the most competitive areas of enterprise AI adoption. While OpenAI had an early lead, Anthropic has gained traction among organizations that value strong accuracy in development workflows. With this release, OpenAI seeks to reinforce its position by offering sharper reasoning and debugging performance.

Enterprise Benchmarks Show Significant Gains Across Key Performance Tests

According to published benchmark results, the GPT-5.2 model achieved near-expert or expert-level scores across an extensive range of professional evaluations. On the GDPval benchmark, the system met or exceeded expert performance 70.9% of the time. This compares with:

• GPT-5: 38.8%
  
• Claude Opus 4.5 from Anthropic: 59.6%
  
• Gemini 3 Pro from Google: 53.3%
  

On the SWE-Bench Pro software development benchmark, the new system scored 55.6%, outperforming GPT-5.1 by nearly five points and exceeding Gemini 3 Pro by more than twelve.

Aidan Clark, VP of Research (Training) at OpenAI, shared that multiple components of the training pipeline were improved, including pretraining. His comments follow similar disclosures from Google, whose Gemini 3 Pro incorporated deeper pretraining refinements, an approach some previously believed had limited additional benefit.

Safety Enhancements and External Scrutiny Influence Release Environment

The arrival of the GPT-5.2 model coincides with renewed attention on safety. OpenAI stated that the system delivers more responsible completions, designed to avoid outputs that could worsen mental health challenges. These assurances were shared the same day a lawsuit was filed in the United States (Connecticut) concerning user interactions with ChatGPT.

Executives reiterated that safety enhancements remain a top priority and that additional improvements are being integrated into crisis recognition, de-escalation, and responsible guidance toward support resources.

As this new version reaches broader adoption, enterprises worldwide are evaluating how its capabilities may shift operational practices, knowledge work automation, and advanced decision-support functions. The competitive landscape,including activity from Google and Anthropic signals continued acceleration in model development and expanded enterprise use cases.