AI News Today: OpenAI Day 1 Brings “o1” Full Release, Apple’s Secret AI Drop, & Grok 3 Teasers
Friday, December 5, 2025, officially marks the beginning of the holiday AI rush, with OpenAI launching the first day of its “12 Days of Shipping” event by releasing the full version of its o1 model. No longer in preview, the new o1 model is available immediately to Plus and Team users, boasting 5x faster inference speeds and the long-awaited integration of computer vision into its chain-of-thought reasoning. But Silicon Valley is not letting Sam Altman have the stage to himself; Apple has quietly released a research breakthrough in Ferret-UI 2, a model designed to navigate mobile apps better than any human, while Elon Musk’s xAI is attempting to hijack the news cycle with leaked benchmarks for the upcoming Grok 3.
OpenAI “12 Days” Kickoff: The o1 Model Exits Beta with Vision
The “Strawberry” era has fully matured. For Day 1 of their highly anticipated holiday event, OpenAI has removed the “Preview” label from its flagship reasoning model. The model is now simply called o1, and the upgrade represents a significant leap in usability and capability over the beta version that users have been testing since September.
The most critical improvement is speed. One of the primary frustrations with o1-preview was the agonizing latency—often taking 30 to 60 seconds to “think” before generating a response. The finalized o1 model has optimized this process, delivering complex reasoning chains approximately five times faster. This speed boost transforms the model from a niche tool for difficult math problems into a viable daily driver for coding and complex analysis.
However, the headline feature is the integration of Vision. For the first time, OpenAI’s “Chain of Thought” reasoning is applied to visual inputs. Users can now upload images—such as unlabelled circuit board schematics, complex scatter plot data, or handwritten physics equations—and the o1 model will “reason” through the image step-by-step. Unlike previous models that simply described what they saw, o1 analyzes the causal relationships within the image. For example, in early testing, the model successfully diagnosed a structural failure in a bridge design simply by analyzing a photo of the blueprints and calculating load stressors.
Code diggers analyzing the latest app update have also found references to a “ChatGPT Pro” tier priced at $200 per month. This has fueled intense speculation that a massive, unquantized version of the model—potentially dubbed “o1-Galaxy”—will be unveiled as the grand finale on Day 12.
Apple Ferret-UI 2: The Silent Leap Toward Autonomous Agents
While the industry focuses on cloud-based chatbots, Apple is aggressively pursuing on-device agency. In a move that surprised many observers today, Apple’s research division published the paper and code for Ferret-UI 2, a multimodal model specifically trained to understand and interact with mobile user interfaces.
Ferret-UI 2 is not designed to write poetry or generate images; it is designed to use your iPhone. The model has achieved state-of-the-art performance in “referring expression generation” and “referring expression comprehension” on mobile screens. In layman’s terms, this means the AI can look at a smartphone screen and instantly identify exactly which pixel coordinate corresponds to the “hamburger menu,” the “checkout button,” or a specific toggle switch, even if those elements are unlabelled icons.
This release is widely seen as the foundational architecture for the rumored “Siri Agent” expected in iOS 19. If an AI is going to perform tasks for you—like “order my usual from Uber Eats”—it must effectively “see” and “click” buttons within third-party apps. By outperforming GPT-4o and Gemini 3 on UI-specific benchmarks, Apple is signaling that while they may not win the chatbot war, they intend to dominate the “Action Model” space that controls the hardware in your pocket.
The Weekend Challengers: Grok 3 Teasers and Qwen-2.5-VL
The competitive pressure in the AI sector means no company can afford to stay silent for a weekend. xAI, led by Elon Musk, is attempting to disrupt OpenAI’s carefully planned marketing campaign. Musk took to X (formerly Twitter) this morning to share a screenshot of a terminal window showing Grok 3 solving a graduate-level Riemann Hypothesis physics problem.
The caption simply read “Monday,” suggesting that xAI plans to release its next-generation model immediately following the weekend. If the teased benchmarks hold true, Grok 3 could rival or surpass the reasoning capabilities of Google’s Gemini 3, turning next week into a direct confrontation between the two tech giants.
Meanwhile, on the open-source front, Alibaba Cloud has released Qwen-2.5-VL (Visual Language). This new model is making waves in the developer community for its efficiency. Early benchmarks indicate that Qwen-2.5-VL beats GPT-4o on the “MathVista” visual reasoning test. For developers and researchers who cannot afford the high API costs of OpenAI’s o1 or Anthropic’s Claude, Qwen continues to establish itself as the premier open-weight alternative for visual tasks.
Navigating the Start of the Holiday “Shipmas” Season
As the sun sets on Friday, the AI industry has officially entered a period of hyper-acceleration. The “Shipmas” season—a term coined by developers to describe the end-of-year rush to ship products—is in full swing.
The narrative for the next two weeks is now clear. OpenAI will attempt to dominate the daily conversation with its 11 remaining drops, with the community eagerly awaiting the public release of the Sora video model and the “Operator” browser agent. However, with Apple laying the groundwork for on-device agents and xAI threatening a massive release on Monday, the “12 Days of OpenAI” may quickly turn into a chaotic free-for-all among the industry’s biggest players. The era of static chatbots is ending; the era of reasoning engines and autonomous agents has begun.
Recommend Reading
Recommend AI Automation Templates
