👾 Google Just Supercharged AI Agents – Meet Gemini 2.5

Gemini 2.5 boosts AI agents with speed and scale — plus OpenAI’s and xAIs deals, Deepgram voice tech, meeting and video tools, and more updates.

Welcome to The Agent Roundup

Here’s your roundup of what’s new and relevant in the world of agent systems and automation. Google rolls out major upgrades to its Gemini models, OpenAI and xAI secure major funding, and Deepgram launches a streamlined API for building real-time voice agents.

Plus, we spotlight new tools for meetings and video models that are redefining what generative AI can do.

Let’s get into it.👇

This week’s topics:

  • Google upgrades its Gemini 2.5 model family

  • New AI tool for building high-quality voice agents

  • Plus AI investments, trending AI tools, community highlights, and more

AI Agent News Roundup

💥 Breakthroughs

Upgrades to Google Gemini Models

Gemini 2.5 hero

Source: Google

Google’s updated Gemini 2.5 models – Pro, Flash, and Flash-Lite – offer enhanced performance for various AI tasks.

Gemini 2.5 Pro is a high-powered model excelling in complex reasoning and coding, delivering performance comparable to top-tier models like o3-pro at a lower cost.

Gemini 2.5 Flash balances speed and efficiency, ideal for real-time applications like chatbots and data extraction, with improved reasoning and a 1-million-token context window.

Gemini 2.5 Flash-Lite, the fastest and most cost-efficient, is optimized for high-volume, latency-sensitive tasks such as translation and UI coding, capable of generating code for a webpage in seconds.

All models support multimodal inputs (text, images, audio) and adjustable "thinking" budgets for cost and performance optimization, with Pro and Flash now available in production-ready versions and Flash-Lite in preview.

📈 Investments

🇺🇸 OpenAI won a $200M U.S. defense contract to deploy advanced AI systems for military missions.

🇺🇸 xAI in talks to raise $4.3B in new equity funding for its AI operations.

🇺🇸 Tesla invited non-employees to try its Robotaxi service, marking a significant milestone in deploying autonomous AI agents in real-world transportation.

🇯🇵 Softbank’s founder, Son, aims to set up a $1T industrial complex in Arizona with an AI and robotics focus, partnering with TSMC.

🇸🇦 Saudi Arabia partners with Replit to launch an Arabic-first version of Replit, bringing AI coding tools to governments, enterprises, and individuals at scale.

AI Meeting Assistant with #1 Noise Cancellation

krisp app with meeting transcription and action items

Krisp automatically records, transcribes, and summarizes your meetings in real time, so you can stay focused on the discussion while never missing a detail. Its AI-powered notes highlight key points and action items, making follow-ups effortless.

Plus, Krisp’s advanced noise cancellation ensures crystal-clear audio by removing background distractions, all without needing extra plugins. Seamlessly compatible with any conferencing app like Zoom, Teams, or Slack, Krisp is the ultimate tool for professionals who want stress-free, efficient meetings.

Join Krisp today and experience smarter meetings with unlimited free transcriptions and easy sharing to keep your team aligned and accountable.

Tool Spotlight

👾 Build High-Quality Voice Agents with Deepgram Voice API

Source: Deepgram

The Deepgram Voice Agent API offers a single, unified solution for building real-time voice agents, combining speech-to-text (STT), LLM processing, and text-to-speech (TTS) in one streamlined API. It eliminates the need for stitching together multiple services, significantly reducing integration time and engineering complexity.

🔑 Key Features & Capabilities

  • Unified WebSocket API: Integrates STT (Nova-3), TTS (Aura-2), and LLMs into one interface for seamless voice interaction.

  • Real-time Conversational Control: Advanced barge-in detection and end-of-turn prediction enable smooth, natural interactions.

  • Flexible Deployment Options: Run in the cloud, your own VPC, or on-prem with support for HIPAA and GDPR compliance.

  • Bring-Your-Own Models: Option to integrate your own LLMs or TTS engines while retaining full orchestration control.

  • Multilingual Support: Expanded transcription capabilities with Nova-3 across multiple languages.

  • Cost-Effective at Scale: Priced at $4.50/hour – significantly cheaper than ElevenLabs and OpenAI alternatives, with discounts when using custom models.

🚀 Built for Developers & Enterprises

  • Simple enough for developers to build fast.

  • Powerful enough for enterprises to maintain control and compliance.

  • Backed by benchmarks showing superior performance in latency, conversational fluidity, and response accuracy.

  • Ideal for teams building production-ready voice agents that require speed, control, and quality without the orchestration hassle.

🎬 MiniMax Hailuo 02: A new AI video model that reaches No. 2 on benchmarks, overtaking Google Veo 3.

🎞️ Midjourney video 1: A new video generation model enabling users to animate any image into 5-second clips.

Community Highlights

More Resources

Blog: In-depth articles on AI workflows and practical strategies for growth
AI Tool Collection: Discover and compare validated AI solutions
Consultancy: Explore AI potential or make your team AI-fit
Agency: AI implementation services to scale your business

See you next time!

Tobias from The Agent Roundup

P.S.: I renamed Agents Made Simple to The Agent Roundup and will transition the newsletter to the domain agentroundup.com.