AI News Today: OpenAI Kicks Off “12 Days” with o1 Full Release, Apple Debuts Ferret-UI 2, & Grok 3 Teasers
Friday, December 5, 2025, marks the official start of the holiday AI season. OpenAI has pulled the trigger on its “12 Days of Shipping” event, delivering a major upgrade to its reasoning capabilities right out of the gate. But they aren’t the only ones making moves today—Apple has quietly dropped a research bombshell that could redefine Siri, and Elon Musk is determined to disrupt the weekend news cycle with Grok 3 teasers.
OpenAI Day 1: “o1” Exits Preview and Gains Sight
The wait for “Strawberry” to mature is over. For Day 1 of their holiday event, OpenAI has officially removed the “Preview” tag from its reasoning model. The model is now simply called o1, and it comes with two massive upgrades that address the biggest complaints from the beta period.
First, Speed. The full o1 model solves complex math and coding problems approximately 5x faster than o1-preview. The agonizing 30-second “thinking” pauses have been optimized down to brief moments of reflection, making the model feel viable for daily workflow rather than just niche problem-solving.
Second, and most importantly, Vision. The o1 model can now apply its “Chain of Thought” reasoning to images.
- Real-world application: You can upload a photo of a complex circuit board, and o1 won’t just guess what it is; it will “think” through the signal path, trace the components step-by-step, and output a diagnostic analysis.
- Availability: The update is live now for ChatGPT Plus and Team users.
However, the code sleuths are already looking ahead. Hidden strings in today’s app update reference a “ChatGPT Pro” tier priced at $200/month, fueling speculation that a massive, unquantized version of o1 (perhaps “o1-Galaxy”) will be the “Day 12” finale.
Apple’s Strategic Drop: Ferret-UI 2 for Mobile Agents
While OpenAI focuses on the cloud, Apple is focusing on your screen. In a surprise research blog post today, Apple unveiled Ferret-UI 2, the second generation of its multimodal model designed specifically to understand mobile interfaces.
Unlike general vision models (like Gemini 3) which look at “photos of the world,” Ferret-UI 2 is trained on iOS and Android screens. It understands what a “hamburger menu,” a “back button,” and a “checkout field” are, even if they aren’t labeled with text.
- Why this matters: This is the missing link for the rumored “Siri Agent” in iOS 19. If an AI is going to “order food for you,” it needs to know where to click. Ferret-UI 2 outperforms GPT-4o significantly in “UI referencing,” proving Apple is building the brain for an on-device agent that actually works.
The Challengers: Grok 3 & Qwen-2.5-VL
The week wouldn’t be complete without a prod from xAI. Elon Musk took to X (formerly Twitter) this morning to share a screenshot of Grok 3 solving a graduate-level physics problem that allegedly stumped Google’s Gemini 3.
- The Tease: Musk captioned the image “Monday,” suggesting that while OpenAI is doing a 12-day slow drip, xAI plans to drop their entire massive model update at the start of next week to steal the momentum.
Meanwhile, in the open-weights world, Alibaba Cloud released Qwen-2.5-VL (Visual Language). Early benchmarks show this model beating GPT-4o on the “MathVista” benchmark. For developers who can’t afford OpenAI’s API costs, Qwen continues to be the most powerful open alternative for visual tasks.
The “Shipmas” Calendar: 11 Days Left
We are only on Day 1. With the reasoning model now fully released, the community expects the next 11 days to bring:
- Sora (Video generation).
- The “Operator” Agent (Browser control).
- Canvas 2.0 (Better coding UI).
The “AI Arms Race” has turned into a daily spectator sport. Check back tomorrow to see what OpenAI drops for Day 2.
Recommend Reading
Recommend AI Automation Templates
