Google Gemini Advances Reasoning While Microsoft Boosts AI Security

Q: What does Google Gemini Advances Reasoning While Microsoft Boosts AI Security cost?

The pricing for Google Gemini Advances Reasoning While Microsoft Boosts AI Security is: Check the official website. Visit the official website for the most up-to-date pricing information and available plans.

The AI landscape is currently experiencing rapid advancements and strategic shifts across major tech players. Google's Gemini 3 Pro Preview has demonstrated exceptional reasoning capabilities, achieving the highest score on the challenging CritPt physics benchmark with 9%, nearly doubling GPT 5.1's 4.9%. Furthermore, Gemini 3 Pro Preview also led the FrontierMath benchmark, solving 9 out of 48 of the hardest Tier 4 math problems for an 18.8% score, surpassing OpenAI's GPT-5.1, GPT-5, and GPT-5 Pro, which all scored 12.5%. This highlights Google DeepMind's strong focus on mathematical reasoning, while other models like Claude 4.5 Haiku and Llama 4 failed to solve any problems on CritPt. Meanwhile, Microsoft is bolstering its cybersecurity defenses, announcing new AI security agents at its Ignite conference on November 21, 2025. These agents are designed to combat AI-enabled hackers, a threat highlighted by a September 2025 espionage campaign. Available free for Copilot Security customers with 365 E5 subscriptions, these tools, like the Phishing Triage Agent, integrate with existing platforms such as Entra, Intune, and Defender to detect and remediate vulnerabilities. In the realm of coding, a comparison between Composer and Claude 4.5 Sonnet revealed their distinct strengths. Composer excels as a coding assistant for multi-file edits and rapid web development, quickly generating working code. Claude 4.5 Sonnet, a general AI model, proved superior for algorithms and system design, offering better explanations and code quality, suggesting a combined approach for optimal results. The development of 3D modeling tools also continues to evolve. Tests in November 2025 compared SAM 3D with Luma AI and Gaussian Splatting. SAM 3D, which generates a 3D model from a single image in seconds, is fast and free but guesses hidden parts and bakes lighting. Luma AI, requiring multiple views and cloud processing, delivered superior realism and captured real backsides, especially for reflective objects. Similarly, Gaussian Splatting, using many images or video frames, offered higher realism and better lighting effects compared to SAM 3D's single-image approach. SAM 3D Objects proved effective for product geometry like shoes and mugs, while SAM 3D Body was better suited for human figures, both running efficiently on an RTX 4070 laptop. Beyond product development, the broader implications of AI are being addressed. The Trump administration has temporarily halted a draft executive order that would have replaced state AI laws with less strict federal rules. In the corporate world, General Motors' first chief AI officer, Barak Turovsky, departed on November 22 after eight months, as GM shifts to integrate AI capabilities directly into its business and product groups rather than maintaining a centralized AI division. Google is also prioritizing AI safety, particularly for users in India. Ahead of the AI Impact Summit 2026 in New Delhi, Google announced initiatives like CodeMender, an AI agent that fixes security flaws, and real-time scam detection on Pixel phones using Gemini Nano. The company has also blocked over 115 million risky apps in India via Google Play Protect and supports digital literacy programs and research with IIT Madras. Finally, a new app has emerged, aiming to help people work alongside AI, fostering collaboration to prevent job displacement.

Key Takeaways

Google's Gemini 3 Pro Preview scored 9% on the CritPt physics benchmark, nearly doubling GPT 5.1's 4.9%, and achieved 18.8% on FrontierMath's hardest problems, surpassing OpenAI's GPT-5.1, GPT-5, and GPT-5 Pro (all 12.5%).
Microsoft launched new AI security agents at its Ignite conference on November 21, 2025, to combat AI-enabled hackers, offering them free to Copilot Security customers with 365 E5 subscriptions.
Claude 4.5 Sonnet excels in AI reasoning, explanations, and generating high-quality code for algorithms and system design, while Composer is faster for multi-file edits and web development.
Luma AI and Gaussian Splatting provide superior realism and detail for 3D models by using multiple images or video, compared to SAM 3D's faster, single-image approach which guesses hidden parts.
SAM 3D Objects is effective for product geometry, and SAM 3D Body accurately captures human figures, both running efficiently on an RTX 4070 laptop.
The Trump administration temporarily halted a draft executive order aimed at replacing state AI regulations with less strict federal rules.
General Motors' first chief AI officer, Barak Turovsky, departed on November 22, as the company moves to decentralize AI capabilities by integrating them directly into business units.
Google is enhancing AI safety for Indian users with initiatives like CodeMender for security flaw fixes, real-time scam detection on Pixel phones via Gemini Nano, and blocking over 115 million risky apps with Google Play Protect.
A new app has been developed to help people work collaboratively with AI, aiming to prevent job displacement by integrating human roles with AI technology.

SAM 3D and Luma AI Battle for Best 3D Tool

This article compares SAM 3D and Luma AI, two tools for creating 3D models. SAM 3D uses one image and is fast, but it guesses unseen parts and bakes lighting. Luma AI uses multiple views, offering better quality and real backsides, especially for reflective objects. Tests on a ceramic fox, mug, and sneaker in November 2025 showed Luma AI scored higher for realism. SAM 3D is free and quicker for a first mesh, while Luma AI needs a capture session and cloud processing but delivers superior final quality.

SAM 3D Objects and Body Models Compared

This article compares two versions of SAM 3D: Objects and Body. SAM 3D Objects works best for items like shoes and mugs, creating clean 3D shapes and textures quickly. It struggles with clear or very thin objects. SAM 3D Body is designed for human figures, accurately capturing poses and clothing, though complex patterns can be tricky. Tests in November 2025 showed Objects excelled for product geometry, while Body was better for human forms. Both models ran efficiently on an RTX 4070 laptop.

SAM 3D and Gaussian Splatting for 3D Models

This article compares SAM 3D and Gaussian Splatting for creating 3D models. SAM 3D uses one image to quickly guess a full 3D object, but it invents hidden parts. Gaussian Splatting uses many images or video frames to reconstruct what the camera actually saw, offering higher realism and better lighting effects. Tests in November 2025 on a mug showed Gaussian Splatting captured more detail. SAM 3D is faster for single images, taking seconds, while Gaussian Splatting requires more input and minutes for processing but delivers superior quality for complex scenes.

Gemini 3 Excels in New Physics AI Test

Google's Gemini 3 Pro Preview has achieved the highest score on a new and very difficult physics research benchmark called CritPt. This benchmark was created by over 60 researchers from many institutions, including Argonne National Laboratory. It features 70 graduate-level physics problems that require deep understanding and are not found online. Gemini 3 scored 9%, almost twice as much as its closest competitor, GPT 5.1, which scored 4.9%. Other major AI models like Claude 4.5 Haiku and Llama 4 failed to solve any problems. This shows Gemini 3's strong reasoning abilities in complex physics.

Gemini 3 Leads AI in Advanced Math Challenge

Google's Gemini 3 Pro Preview has achieved the top scores on the FrontierMath benchmark, which tests AI models on very difficult math problems. These problems range from undergraduate to research-level and are unpublished, ensuring models rely on true reasoning. On the hardest Tier 4 problems, Gemini 3 Pro Preview scored 18.8%, solving 9 out of 48. This is significantly higher than OpenAI's GPT-5.1, GPT-5, and GPT-5 Pro, which all scored 12.5%. Gemini 3 also outperformed competitors on easier math problems. These results highlight Google DeepMind's strong focus on mathematical reasoning.

Composer and Claude 4.5 Sonnet Coding Showdown

This article compares Composer and Claude 4.5 Sonnet for practical coding tasks. Composer is a coding assistant good at multi-file edits and project-wide changes, quickly creating working code. Claude 4.5 is a general AI model strong in reasoning, explanations, and generating structured, well-commented code. The author tested both on 15 tasks, including web development and algorithms, using a MacBook Pro M3. While Composer was faster for web development, Claude 4.5 excelled in algorithms and system design with better explanations and code quality. The best approach is to use Composer for initial drafts and Claude 4.5 for review and refinement.

New App Helps People Work With AI

A new company has created an app designed to help people work alongside artificial intelligence. The goal is to integrate human jobs with AI technology, hoping to prevent job displacement caused by AI advancements. NBC News' Yasmin Vossoughian reported on this new innovation on November 22, 2025. This app aims to keep people in their roles by enabling them to collaborate with AI.

Personal Journey From Hawaii to Colorado

This article shares a personal story about the author's life journey. The author was born in Hawaii and moved many times before settling in Golden for high school in the late 1980s. They later moved to Denver for college, attending Metro and The Art Institute of Colorado. The author also worked in various non-art jobs.

Trump Halts Federal AI Regulation Order

The Trump administration has temporarily put aside a draft executive order related to artificial intelligence. This order aimed to replace state laws on AI with less strict federal rules. The decision means the federal government will not move forward with these specific AI guidelines for now.

Microsoft Launches AI Security Agents Against Hackers

Microsoft announced new AI security agents at its Ignite conference on November 21, 2025, to help businesses fight AI-enabled hackers. These agents are built into Microsoft's security dashboards and are free for Copilot Security customers with 365 E5 subscriptions. Hackers are now using AI for sophisticated attacks, as seen in a September 2025 espionage campaign. Microsoft's new agents, like the Phishing Triage Agent, help detect and fix vulnerabilities, integrating into tools like Entra, Intune, and Defender to protect against evolving cyber threats.

GM AI Chief Barak Turovsky Departs Company

Barak Turovsky, General Motors' first chief artificial intelligence officer, has left the company after eight months. GM confirmed his departure on November 22, stating his team will now report to manufacturing engineering. This move is part of GM's plan to integrate AI capabilities directly into business and product groups instead of keeping AI centralized. Turovsky previously worked at Google and his departure follows that of Dave Richardson, GM's senior vice president of software and services engineering, on October 31.

Google Boosts AI Safety for Indian Users

Google is creating safe and trusted AI to protect users in India, focusing on cybersecurity and digital literacy. The company announced initiatives ahead of the AI Impact Summit 2026 in New Delhi, the first global AI summit in the Global South. Google introduced CodeMender, an AI agent that fixes security flaws, and real-time scam detection on Pixel phones using Gemini Nano. Google Play Protect has blocked over 115 million risky apps in India, and the company supports digital literacy programs and research with IIT Madras to advance AI safety.