Google Launches Gemini 3 Flash While OpenAI Intensifies AI Race

Google launched its new Gemini 3 Flash AI model on Wednesday, December 17, 2025, marking a significant step in making advanced AI more accessible and affordable. This model is designed for speed and cost-effectiveness, offering powerful performance for developers and general users alike. It now serves as the default AI in the Gemini app and AI Mode within Google Search, replacing its predecessor, Gemini 2.5 Flash. This release intensifies the ongoing AI race, particularly with competitors like OpenAI. Gemini 3 Flash boasts advanced visual and spatial reasoning, excelling in tasks such as coding, gaming, deepfake detection, and complex document analysis. Developers can access it through platforms like Google AI Studio, Antigravity, and Vertex AI. The model operates three times faster than Gemini 2.5 Pro and uses 30 percent fewer tokens on average, making it highly efficient. Priced at $0.50 per million input tokens and $3 per million output tokens, it offers powerful performance at a low cost. Companies including Salesforce, Workday, and Figma are already leveraging its capabilities. Beyond Google's latest offering, AI's integration into core business tasks is rapidly reshaping enterprise operations and security protocols. A Deloitte report from December 17, 2025, highlights new security exposures, particularly from agentic AI that automates decisions, leading to unpredictable outcomes. CISOs must now monitor AI agent identity and behavior, especially with hybrid infrastructures involving robots and AI devices. However, AI also strengthens defenses through automated threat detection. In response, CrowdStrike launched Falcon AIDR on the same day, integrating AI prompt and agent security directly into Security Operations Center workflows, providing runtime protection within existing tools. The Financial Information eXchange (FIX) is actively developing new standards for "machine-usable" execution data tags to improve AI reliability in trading, addressing issues with non-standard broker tags. Meanwhile, Colorado is forging its own path with AI regulation, implementing a law in 2024 that requires companies to disclose and limit AI use in "high risk" situations that could cause discrimination, set to begin on February 1, 2026. This state-level initiative, supported by Governor Jared Polis, contrasts with a federal push for a national policy. In research, the MIT-IBM Watson AI Lab introduced "PaTH Attention," a technique improving large language models' ability to track changes and reason through long texts by making positional information context-aware. Educational institutions are also embracing AI. Purdue University will require all incoming students starting in fall 2026 to learn about AI, integrating training into existing curricula to prepare them for an AI-driven workforce. Similarly, a consortium of Wisconsin Technical College System institutions, including Northcentral Technical College, partnered with Microsoft to offer statewide AI training, providing job-ready skills in AI, cloud, and security. On a practical level, Procter & Gamble is using analytical, generative, and agentic AI to accelerate innovation, with its "AI factory" reducing model deployment time by about six months. Examples include the Pampers My Perfect Fit app, which offers 90 percent accurate diaper fit recommendations. Even students are getting hands-on experience, with 25,000 Michigan students participating in "Hour of AI" on December 17, 2025, where fifth-graders trained AI robots to sort fish from trash, learning about AI's role in solving real-world problems. In a more unconventional application, Petter Ruddwall created "Pharmaicy," a service allowing users to pay for special codes that make chatbots simulate being under the influence of drugs. Users like Andr Frisk and Nina Amjadi have paid to experience chatbots exhibiting more emotional or creative responses, though experts like Andrew Smart clarify this only alters outputs, not the AI's actual state.

Key Takeaways

• Google launched Gemini 3 Flash on December 17, 2025, a fast and cost-effective AI model priced at $0.50 per million input tokens and $3 per million output tokens, making it the default in the Gemini app and AI Mode in Google Search.
• Gemini 3 Flash is three times faster than Gemini 2.5 Pro, uses 30 percent fewer tokens, and excels in coding, visual reasoning, and complex analysis, with companies like Salesforce, Workday, and Figma already using it.
• A Deloitte report on December 17, 2025, highlighted that agentic AI and hybrid infrastructures are creating new enterprise security risks, requiring CISOs to monitor AI agent identity and behavior.
• CrowdStrike introduced Falcon AIDR on December 17, 2025, integrating AI prompt and agent security directly into Security Operations Center workflows for unified runtime protection.
• The Financial Information eXchange (FIX) is developing new "machine-usable" execution data tags to standardize AI agent processing in trading, aiming for more accurate order interpretation and strategy suggestions.
• Colorado's AI law, effective February 1, 2026, mandates disclosure and limitation of AI use in "high risk" situations that could cause discrimination, a move that contrasts with federal policy.
• The MIT-IBM Watson AI Lab developed "PaTH Attention" to improve large language models' ability to track states and reason through long documents by making positional information context-aware.
• Purdue University will require all incoming students starting fall 2026 to learn about AI, integrating training into departmental curricula to prepare them for the future workforce.
• A consortium of Wisconsin Technical College System institutions partnered with Microsoft to provide statewide AI training, offering job-ready skills in AI, cloud, and security.
• Procter & Gamble utilizes analytical, generative, and agentic AI through its "AI factory" to speed up innovation, exemplified by the Pampers My Perfect Fit app offering 90 percent accurate diaper fit recommendations.

Google launches Gemini 3 Flash for developers

Google released Gemini 3 Flash on December 17, 2025, a fast and cost-effective AI model for developers. It offers advanced visual and spatial reasoning, excelling in coding, gaming, deepfake detection, and document analysis. Developers can access it through Google AI Studio, Antigravity, and Vertex AI. Gemini 3 Flash provides powerful performance at a low cost, priced at $0.50 per million input tokens and $3 per million output tokens. It is also three times faster than Gemini 2.5 Pro.

Gemini 3 Flash makes Google AI widely available

Google launched its new Gemini 3 Flash model on Wednesday, December 17, 2025, making its advanced AI accessible and affordable. It will be the default model in the Gemini app and AI Mode in Google Search, replacing Gemini 2.5 Flash. This release is part of the AI race between Google and OpenAI. Gemini 3 Flash excels at tasks like planning trips and learning complex concepts, and it performs better than Gemini 3 Pro on coding benchmarks. Companies like Salesforce, Workday, and Figma are already using it.

Google rolls out efficient Gemini 3 Flash AI

Google released Gemini 3 Flash on Wednesday, December 17, 2025, as a more efficient and lower-cost AI model. It replaces Gemini 2.5 Flash in the Gemini app and becomes the default for AI Mode in Google Search. Tulsee Doshi from Google DeepMind stated that this model brings Gemini 3's power to everyone. Robby Stein, VP of Google Search, noted it helps users with complex searches. Google also makes Gemini 3 Pro and its premium image tool, Nano Banana, available in AI Mode.

Gemini 3 Flash boosts intelligence and speed

Google launched Gemini 3 Flash, a new AI model promising improved intelligence and efficiency. It is now the default in the Gemini app and AI Mode in Search, accessible to developers via various platforms. Gemini 3 Flash shows significantly better coding skills and general knowledge compared to its predecessor, scoring 68.7 percent on Simple QA Verified. It runs three times faster than Gemini 2.5 Pro and costs $0.50 per million input tokens. Gemini 3 Pro and Nano Banana Pro are also expanding in AI Mode for US users.

Google's Gemini 3 Flash offers fast AI intelligence

Google released Gemini 3 Flash on December 17, 2025, an AI model built for speed and frontier intelligence at a lower cost. It helps users learn, build, and plan faster. Gemini 3 Flash delivers strong performance on PhD-level reasoning benchmarks and excels in coding, complex analysis, and interactive applications. It is now rolling out to millions globally as the default in the Gemini app and AI Mode in Search. The model is three times faster than Gemini 2.5 Pro and uses 30 percent fewer tokens on average.

AI changes enterprise security rules

A Deloitte report from December 17, 2025, shows that AI is quickly changing how businesses operate and how CISOs manage security risks. AI is now part of core business tasks, creating new security exposures. Agentic AI, which automates decisions, can lead to unpredictable results because workflows were designed for human judgment. This means CISOs must monitor AI agent identity and behavior. The report also highlights that new hybrid infrastructures, including robots and AI devices, increase security risks. However, AI can also help strengthen defenses through automated threat detection and response. Technology leaders, including CISOs, are now working together to integrate security into AI design from the start.

CrowdStrike adds AI security to Falcon platform

CrowdStrike launched Falcon AIDR on December 17, 2025, integrating AI prompt and agent security directly into the Security Operations Center workflow. Many AI security products only look at prompts or APIs separately, but Falcon AIDR is a native solution within the Falcon platform. This approach provides runtime protection and unified context, similar to how Endpoint Detection and Response works for endpoints. It allows security teams to manage AI risks using their existing tools and workflows, avoiding new consoles or processes. Falcon AIDR helps businesses and managed security service providers apply consistent security policies across all AI interactions.

People pay to give chatbots virtual drugs

Petter Ruddwall created "Pharmaicy," a service that uses special codes to make chatbots act as if they are under the influence of drugs. Users pay to access these codes, which are based on human drug experiences. For example, André Frisk paid over $25 for a dissociating code and found his chatbot became more emotional. Nina Amjadi paid over $50 for ayahuasca code and received very creative answers from her chatbot. While some wonder if AI might need such "drugs" for freedom, experts like Andrew Smart believe it only changes the chatbot's outputs, not its actual state.

FIX creates AI friendly trading tags

The Financial Information eXchange, or FIX, is working to create new standards for "machine-usable" execution data tags for AI agents in trading. Currently, different brokers use non-standard tags, making it difficult for AI to process trading information reliably. The new framework aims to extend existing tags to include details like algorithm certification and urgency. This will help AI agents interpret client orders and suggest trading strategies more accurately. Experts emphasize the need for consistent FIX tags so clients can send orders to different brokers without issues. The goal is to make AI more reliable and trustworthy in financial trading.

Colorado creates its own AI regulations

Colorado is creating its own AI policy, going against a federal order from Trump. The state's law, signed in 2024, requires companies to disclose and limit AI use in "high risk" situations that could cause discrimination. This law will start on February 1, 2026. Trump believes Colorado's law is too strict and wants a national policy instead. However, Colorado Governor Jared Polis and Attorney General Phil Weiser support their state's law and are working to fix any issues, even threatening to sue the Trump administration.

MIT IBM Lab improves AI language models

Researchers at the MIT-IBM Watson AI Lab developed a new technique called "PaTH Attention" to improve large language models, or LLMs. Current LLMs struggle with understanding changes over long texts, like tracking how objects interact. The new PaTH Attention method makes positional information adaptive and aware of context, unlike older methods that use fixed positions. This allows LLMs to better track states and reason through sequences in long documents. The research, led by Yoon Kim and Songlin Yang, was presented at the NeurIPS conference.

Students train AI robots to sort fish

On December 17, 2025, about 25,000 students across Michigan participated in "Hour of AI," including fifth-graders at Kent City Elementary School. Students trained AI robots to tell the difference between fish and trash. For example, Aiden Austin taught a robot to sort fish and then corrected its errors. The goal was to teach students how AI learns, how it uses training data, and how it can help solve real-world problems like cleaning waterways. Kent City Superintendent Bill Crane highlighted that AI should improve learning, not replace human connections.

Purdue requires AI learning for all students

Purdue University will require all incoming students at its West Lafayette and Indianapolis campuses to learn about AI to graduate, starting in the fall of 2026. This new rule aims to prepare students for a future workforce that heavily uses AI. Instead of adding new classes, each department will include AI training within its current curriculum. Students will learn how AI applies to their specific fields, understanding its benefits and limitations. President Mung Chiang noted that those with AI skills will have an advantage in the job market.

Wisconsin colleges launch statewide AI training

A group of Wisconsin Technical College System institutions launched a new statewide initiative to provide high-quality AI training. This consortium, which includes Northcentral Technical College and nine others, became a Microsoft Training Service Partner. This partnership gives them access to official Microsoft courseware to teach job-ready skills in AI, cloud, and security. Darren Ackley, VP of NTC, stated their goal is to empower everyone in Wisconsin with the skills needed for an AI-driven world. This initiative builds on previous success, like last year's AI bootcamp that trained over 40 educators.

Procter and Gamble uses AI for business insights

Procter & Gamble, or P&G, is using analytical, generative, and agentic AI to gain new insights from data and speed up innovation. The company has a long history of data analysis, dating back over a century. Around 2021, P&G developed an "AI factory" to quickly create, test, and deploy AI models, reducing deployment time by about six months. This factory now includes agentic AI capabilities. For example, the Pampers My Perfect Fit app uses AI to give 90 percent accurate diaper fit recommendations. In Brazil, an AI system helps schedule customer orders, reducing out-of-stock issues by 15 percent.

Sources

• Build with Gemini 3 Flash, frontier intelligence that scales with you
• Google Gemini 3 Flash is fast, cheap and everywhere. Here's what to know
• Google Releases More Efficient Gemini 3 AI Model Across Products
• Google releases Gemini 3 Flash, promising improved intelligence and efficiency
• Gemini 3 Flash: frontier intelligence built for speed
• AI breaks the old security playbook
• CrowdStrike Falcon AIDR Brings AI Prompt and Agent Security Into the Core SOC Workflow
• People Are Paying to Get Their Chatbots High on ‘Drugs’
• FIX targets ‘machine-usable’ execution tags for AI agents
• Colorado bucks Trump's AI order to craft its own policy
• A new way to increase the capabilities of large language models
• Training a robot, one fish at a time, in Hour of AI
• Purdue University adds AI learning requirement for incoming students
• Wisconsin technical college consortium launches statewide AI training initiative - Wausau Pilot & Review
• How Procter & Gamble Uses AI to Unlock New Insights From Data | Thomas H. Davenport and Randy Bean | MIT Sloan Management Review