Gemini CLI - A new coding agent from Google
- 🚀 Google has launched Gemini CLI, a free and open-source AI agent that brings the power of Gemini directly to developers’ terminals, offering seamless AI assistance for coding, research, and workflow automation.
- 💡 Gemini CLI provides unmatched free usage limits—up to 60 model requests per minute and 1,000 per day—making advanced AI accessible to individual developers at no charge.
- 🔧 The tool is highly extensible and open source (Apache 2.0), supporting custom prompts, Google Search grounding, and integration with the Model Context Protocol (MCP) for tailored workflows.
- 🤝 Gemini CLI shares its technology with Gemini Code Assist, enabling prompt-driven, multi-step AI coding support in both the terminal and VS Code.
- 🛠️ Getting started is easy: install Gemini CLI and log in with your Google account to unlock powerful, context-aware AI in your command line.
- This certainly has the feel of inspired by Claude Code
Google Cloud donates A2A to Linux Foundation
Today at Open Source Summit North America, the Linux Foundation announced a new independent initiative, in partnership with AWS, Cisco, Google, Microsoft, Salesforce, SAP, and ServiceNow, to host the Agent2Agent (A2A) protocol—a web-standard approach donated by Google, inclusive of SDKs and dev tools . The A2A protocol empowers AI agents to discover each other, securely exchange context, and coordinate complex actions across any platform—eliminating silos and unlocking multi-agent innovation . Backed by over 100 companies, including AWS and Cisco as new validators, A2A is a major step toward interoperable AI ecosystems . Now under the neutral governance of the Linux Foundation, the project ensures open collaboration, IP stewardship, and long-term scalability—fueling the next era of agent-powered solutions across industries .
Search with Voice chats using Gemini
Google is piloting Search Live within its AI Mode (U.S. only, via Google Labs) on Android and iOS. Users can now have real-time voice conversations with Gemini AI—ask questions, get audio replies, and keep chatting naturally. Visual inputs through the camera are supported too (though not yet live in the actual chat). The feature displays relevant links, keeps transcripts, and saves everything in your AI Mode history. Conversations even persist when you switch apps. This is the latest evolution in Google’s shift from traditional search to an interactive, multimedia-centric experience—joining similar efforts from OpenAI, Anthropic, and Apple
Gemini 2.5 Pro and Flash from Google go GA
Google has officially advanced its Gemini 2.5 model lineup as of June 17, 2025: both Gemini 2.5 Pro and Gemini 2.5 Flash are now stable and generally available, while Gemini 2.5 Flash‑Lite enters public preview. Pro offers top-tier reasoning, multimodal understanding, coding capabilities, and handles up to a 1 million-token context—ideal for complex, mission-critical tasks. Flash strikes a balance of speed, cost-efficiency, and robust reasoning, with simplified pricing at $0.30 per million input and $2.50 per million output tokens. Flash‑Lite is the fastest and most economical option, optimized for high-throughput tasks like translation and classification, with reasoning off by default and support for tool use. All three models share the million-token context window, adaptive “thinking” control, and grounding via Google Search, code execution, function-calling, and multimodality—all accessible via Gemini app, AI Studio, Vertex AI, and more.
Here are some key features across the Gemini 2.5 models:
- Hybrid Reasoning Models: Designed to provide excellent performance while being efficient in terms of cost and speed.
- “Thinking” Capabilities: Models can reason through their thoughts before responding, which leads to improved accuracy. Developers have control over the “thinking budget” to balance performance and cost.
- Native Multimodality: Understands and processes inputs across various modalities including text, images, audio, and video.
- Long Context Window: Features a 1 million-token context length, allowing them to comprehend vast datasets and handle complex problems from different information sources.
- Tool Integration: Can connect to tools like Google Search and code execution.
Gemini 2.5 Pro:
- Most Advanced Model: Excels at coding and highly complex tasks.
- Enhanced Reasoning: State-of-the-art in key math and science benchmarks.
- Advanced Coding: Capable of generating code for web development tasks and creating interactive simulations.
Gemini 2.5 Flash:
- Fast Performance: Optimized for everyday tasks and large-scale processing.
- Cost-Efficient: Balances price and performance.
- Live API Native Audio: Offers high-quality, natural conversational audio outputs with enhanced voice quality and adaptability, including features like Proactive Audio and Affective Dialog.
Gemini 2.5 Flash-Lite:
- Most Cost-Efficient and Fastest: Designed for high-volume, latency-sensitive tasks like translation and classification.
- Higher Quality: Outperforms 2.0 Flash-Lite on coding, math, science, reasoning, and multimodal benchmarks.
- Lower Latency: Offers lower latency compared to 2.0 Flash-Lite and 2.0 Flash for a broad range of prompts.
A professional video commercial made for 50 bucks
In Brazil, the production of a professional-quality 1-minute advertising video typically incurs costs exceeding R$100,000 reais ($17,543 dollars). The commercial created for Ulianopolis City Hall in Brazil was made with VEO 3, spending only R$300 reais ($52 dollars) in VEO 3 credits. This is a major disruption to traditional commercial production, suggesting advertising and creative agencies are under threat as models start to produce high-quality, native-language AI output, underscoring the imminent shift in media production workflows.
AlphaEvolve - An agent for designing advanced algorithms
Google DeepMind’s AlphaEvolve project uses large language models to invent algorithms beyond human expertise. The project combines coding skills, algorithm testing, and an evolutionary method to produce new designs. AlphaEvolve has generated more efficient algorithms for various computations and real-world problems, demonstrating the potential for AI to come up with novel ideas through experimentation and evaluation.
Google releases Gemini 2.5 Pro -an improved coder
Early access to Gemini 2.5 Pro Preview (I/O edition) is now available. This updated version significantly improves coding capabilities, particularly for building interactive web apps, and is now available in the Gemini API, Google AI Studio, Vertex AI, and the Gemini app.