The artificial intelligence landscape witnessed a significant escalation in competition as Google and OpenAI simultaneously unveiled major advancements in their foundational models and agentic capabilities. On Thursday, Google launched a “reimagined” version of its research agent, Gemini Deep Research, powered by its cutting-edge Gemini 3 Pro model. This new agent extends beyond mere research report generation, enabling developers to integrate Google’s state-of-the-art research functionalities directly into their applications through the new Interactions API, a pivotal tool for the burgeoning agentic AI era. Gemini Deep Research is engineered to synthesize vast quantities of information and manage extensive context within prompts, catering to diverse tasks from due diligence to drug toxicity safety research. Google also announced future integration of this deep research agent into its core services, including Google Search, Google Finance, the Gemini App, and NotebookLM, signaling a move towards a future where AI agents rather than humans perform direct information retrieval.

A key attribute of Deep Research is its reliance on Gemini 3 Pro’s status as Google's “most factual” model, specifically trained to mitigate AI hallucinations. These hallucinations, where large language models (LLMs) generate false information, pose a critical challenge for complex, long-running agentic tasks involving numerous autonomous decisions. The more choices an LLM makes, the higher the probability that even a single hallucinated decision could compromise the entire output. To substantiate its claims, Google introduced a new benchmark called DeepSearchQA, designed to test agents on intricate, multi-step information-seeking tasks, which it has open-sourced. Additionally, Google evaluated Deep Research on Humanity’s Last Exam, an independent benchmark for general knowledge with highly niche tasks, and BrowserComp, a benchmark for browser-based agentic tasks. While Google’s new agent outperformed competitors on DeepSearchQA and Humanity’s Last Exam, OpenAI’s ChatGPT 5 Pro proved a surprisingly close second, even slightly surpassing Google on BrowserComp.

Coinciding with Google's announcement, OpenAI launched its highly anticipated GPT-5.2, codenamed Garlic, on the same day. Positioned as its most advanced model yet, GPT-5.2 is designed for both developers and everyday professional use. It is available to ChatGPT paid users and developers via the API in three distinct versions: Instant, optimized for speed in routine queries like information-seeking, writing, and translation; Thinking, excelling in complex structured tasks such as coding, long document analysis, math, and planning; and Pro, the premium model for maximum accuracy and reliability in challenging problems. OpenAI’s chief product officer, Fidji Simo, highlighted GPT-5.2’s enhanced capabilities in creating spreadsheets, building presentations, writing code, perceiving images, understanding long context, utilizing tools, and orchestrating complex, multi-step projects.

GPT-5.2’s release intensifies the arms race with Google’s Gemini 3, which currently leads the LMArena leaderboard across most benchmarks, with the exception of coding, where Anthropic’s Claude Opus-4.5 maintains an advantage. The launch follows an internal “code red” memo from OpenAI CEO Sam Altman, reportedly prompted by declining ChatGPT traffic and concerns over losing consumer market share to Google. This strategic shift prioritized improving the ChatGPT experience, even delaying other commitments like introducing advertisements. GPT-5.2 thus represents OpenAI's aggressive move to reclaim leadership, focusing on bolstering its enterprise opportunities by targeting developers and the tooling ecosystem to establish itself as the default foundation for AI-powered applications. Recent data from OpenAI indicates a dramatic surge in enterprise usage of its AI tools over the past year.

OpenAI asserts that GPT-5.2 establishes new benchmark scores in coding, math, science, vision, long-context reasoning, and tool use, promising more dependable agentic workflows, production-grade code, and sophisticated systems operating across large contexts and real-world data. These capabilities directly challenge Gemini 3’s Deep Think mode, Google's major reasoning advancement targeting math, logic, and science. On OpenAI’s internal benchmark chart, GPT-5.2 Thinking demonstrably edges out Gemini 3 and Anthropic’s Claude Opus 4.5 in nearly every listed reasoning test, encompassing real-world software engineering tasks (SWE-Bench Pro), doctoral-level science knowledge (GPQA Diamond), and abstract reasoning and pattern discovery (ARC-AGI suites). Research lead Aidan Clark emphasized that improved math scores signify a model’s ability to follow multi-step logic, maintain numerical consistency, and prevent subtle errors from compounding, which are critical properties for financial modeling, forecasting, and data analysis. Product lead Max Schwarzer added that GPT-5.2