🗞️ April 2025 AI Roundup — What You Shouldn’t Miss

🤖 Introducing GPT-4.1 in the API: Improved reasoning and instruction-following for developers.
https://openai.com/index/gpt-4-1/

⚙️ Introducing o3 and o4-mini: Two new model tiers for broad capabilities and lightweight edge inference.
https://openai.com/index/introducing-o3-and-o4-mini/

🧠 X: Tweet “ChatGPT memory upgrade”: Demo of persistent chat history on X.
https://x.com/TheDaveStack/status/1910644388381831442

📰 The latest AI news we announced in March: A recap that set the stage for April’s follow-ups.
https://blog.google/technology/ai/google-ai-updates-march-2025/

🔍 Bringing multimodal search to AI Mode: Combine text, images, and more in a single Google Search query.
https://blog.google/products/search/ai-mode-multimodal-search/

✨ New ways to interact with information in AI Mode: Smarter filters and context-aware suggestions launched in late April.
https://blog.google/products/search/ai-mode-updates-may-2025/

🔒 Google announces Sec-Gemini v1: An experimental cybersecurity model for threat detection and mitigation.
https://security.googleblog.com/2025/04/google-launches-sec-gemini-v1-new.html

🐬 DolphinGemma: Using Google AI to decode dolphin communication patterns.
https://blog.google/technology/ai/dolphingemma/

📊 Deep Research on Gemini 2.5 Pro Experimental: Unlock advanced data-analysis workflows for early adopters.
https://blog.google/products/gemini/deep-research-gemini-2-5-pro-experimental/

🤝 Announcing the Agent2Agent Protocol (A2A): A new standard for AI agent interoperability.
https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability

🚄 Gemini 2.5 Flash & Pro, Live API, and Veo 2: Major API updates for developers.
https://developers.googleblog.com/en/gemini-2-5-flash-pro-live-api-veo-2-gemini-api

🎥 Generate videos in Gemini and Whisk with Veo 2: Video generation goes live in Gemini’s toolkit.
https://blog.google/products/gemini/video-generation/

🚗 Introducing Mobility AI: Research on optimizing urban transportation with AI.
https://research.google/blog/introducing-mobility-ai-advancing-urban-transportation/

🎓 Anthropic Education Report: How university students are using Claude in their studies.
https://www.anthropic.com/news/anthropic-education-report-how-university-students-use-claude

📦 Introducing the Max Plan: Anthropic’s new tiered service offering.
https://www.anthropic.com/news/max-plan

🔬 Claude takes research to new places: Case studies on Claude’s research-assist capabilities.
https://www.anthropic.com/news/research

⚠️ Our Approach to Understanding and Addressing AI Harms: Framework for mitigating AI’s potential harms.
https://www.anthropic.com/news/our-approach-to-understanding-and-addressing-ai-harms

🛡️ Detecting and Countering Malicious Uses of Claude: March 2025 update on safety measures.
https://www.anthropic.com/news/detecting-and-countering-malicious-uses-of-claude-march-2025

📈 Anthropic Economic Index: Report on AI’s impact on software development economics.
https://www.anthropic.com/research/impact-software-development

💼 Introducing the Anthropic Economic Advisory Council: New panel to guide AI investment and policy.
https://www.anthropic.com/news/introducing-the-anthropic-economic-advisory-council

⚖️ Securing America’s Compute Advantage: Anthropic’s position on the Diffusion Rule export controls.
https://www.anthropic.com/news/securing-america-s-compute-advantage-anthropic-s-position-on-the-diffusion-rule

❓ What is DeepSeek AI?: Unveiling China’s open-source AI revolution.
https://deepseek.ai/blog/what-is-deepseek-ai

⚔️ Llama 4 vs DeepSeek AI: A comprehensive head-to-head comparison.
https://deepseek.ai/blog/llama-4-vs-deepseek

🔄 DeepSeek-R2: China’s Bold Answer to the AI Race: Resource-efficient model launch details.
https://deepseek.ai/blog/deepseek-r2-ai-model-launch-2025

💁 Your AI Companion: Microsoft’s vision for a personalized, multi-modal assistant in your daily workflow.
https://blogs.microsoft.com/blog/2025/04/04/your-ai-companion/

🎂 One year of Phi: Celebrating a year of small language models making big efficiency leaps.
https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/

⚖️ Evaluating RAG with LLM as a Judge: New toolkit for retrieval-augmented generation that self-evaluates.
https://mistral.ai/news/llm-as-rag-judge

🌄 The Llama 4 herd: Kicking off a new era of natively multimodal AI.
https://ai.meta.com/blog/llama-4-multimodal-intelligence/

🤖 Devin 2.0: The latest iteration of the Devin agent platform, with richer reasoning and planning controls.👉 Read more

🖼️ V7 Alpha: Early access to Midjourney’s Version 7, with faster generation and finer style controls.
https://www.midjourney.com/updates/v7-alpha

🏰 Introducing Thunderforge: AI for American Defense: New Scale AI tooling aimed at defense analysis and decision-support.
https://scale.com/blog/thunderforge-ai-for-american-defense

👁️‍🗨️ Cogito v1 Preview & Introducing IDA: A path toward general superintelligence with the Intelligent Decision Assistant.
https://www.deepcogito.com/research/cogito-v1-preview

🐋 Introducing Docker Model Runner: A local runtime to build and run GenAI models on your machine.
https://www.docker.com/blog/introducing-docker-model-runner/

📝 DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level: A permissively licensed 14B-parameter code model.
https://pretty-radio-b75.notion.site/DeepCoder-A-Fully-Open-Source-14B-Coder-at-O3-mini-Level-1cf81902c14680b3bee5eb349a512a51