In partnership with

Good morning. It’s Wednesday, April 16th.

On this day in tech history: 1972: Apollo 16 launched toward the Moon with astronauts John Young, Charles Duke, and Ken Mattingly aboard.

In today’s email:

• OpenAI Updates

• Google’s Veo 2 Released

• New Claude Features

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with WRITER}

You’ve heard the hype. Now it’s time for results

After two years of siloed experiments, proofs of concept that fail to scale, and disappointing ROI, most enterprises are stuck. AI isn't transforming their organizations — it’s adding complexity, friction, and frustration.

But Writer customers are seeing a positive impact across their companies. Our end-to-end approach is delivering adoption and ROI at scale. Now, we’re applying that same platform and technology to bring agentic AI to the enterprise.

This isn’t just another hype train that doesn’t deliver. The AI you were promised is finally here — and it’s going to change the way enterprises operate.

See real agentic workflows in action, hear success stories from our beta testers, and learn how to align your IT and business teams.

Watch the product showcase now

OpenAI Updates (So Far): GPT-4.1 Release, New Social Feed, and Looser Guardrails

OpenAI has been rolling out a series of updates this week across its models, tools, and internal processes. Here’s a concise rundown of the key developments:

GPT-4.1 models expand context window, improve code handling: OpenAI has released three new models—GPT-4.1, 4.1 mini, and 4.1 nano—exclusively through its API. These models offer improved coding reliability, faster outputs, and lower costs compared to GPT-4o. All three support context windows of up to one million tokens. GPT-4.1 outperforms GPT-4o on software engineering benchmarks and long-context reasoning, though performance declines on full-length inputs. Smaller variants prioritize speed and affordability; GPT-4.1 mini is 83% cheaper than GPT-4o. Pricing starts at $0.10 per million input tokens, and developers can access GPT-4.1 for free via Windsurf’s platform for a limited time.

OpenAI is also retiring GPT-4.5—its most compute-intensive model—by July, citing high costs and shifting focus to scalable models. Sam Altman also acknowledged that OpenAI's GPT model names have been confusing, promising a fix by summer.

ChatGPT Image Library Rolled Out: A new “Library” tab lets users view and manage their AI-generated images in grid format. Available on Free, Plus, and Pro plans across iOS and web, it also includes a shortcut for generating new content.

Early-Stage Social Platform in Testing: OpenAI is prototyping a social feed focused on ChatGPT-generated images. It’s unclear if this will launch as a standalone app or live within ChatGPT. The effort may serve both user engagement and real-time data collection.

Nonprofit Advisory Group Formed: Four advisors—Dolores Huerta, Monica Lozano, Dr. Robert Ross, and Jack Oliver—have been appointed to help guide OpenAI’s nonprofit initiatives. They’ll deliver community-informed recommendations to the board within 90 days.

New Models for Scientific Reasoning: OpenAI is piloting o3 and o4-mini models built for hypothesis generation and cross-domain experimentation. They power ChatGPT’s Deep Research tool and support use cases like fusion research and plastic recycling. Enterprise pricing may reach $20,000/month.

Safety Rules May Ease Due to Competition: OpenAI has updated its Preparedness Framework, stating it may lower internal safety thresholds if competitors release high-risk AI systems without comparable safeguards. It’s also scaling automated testing and will release new Capabilities and Safeguards Reports alongside manual reviews. While claiming it will remain “more protective,” insiders cite rushed evaluations.

Context.ai Team Joins OpenAI:  The team behind GV-backed evaluation startup Context.ai has been acqui-hired. Co-founders Henry Scott-Green and Alex Gamble will work on internal model transparency tools. Context.ai’s original product has been discontinued.

Google Rolls Out Video Features in Gemini and Seeks Research Scientist for Post-AGI Work

Gemini Advanced users can now generate high-resolution, eight-second videos using the Veo 2 model. By inputting text prompts, users can create videos with fluid character movement and cinematic realism. This feature is available to Google One AI Premium subscribers and includes Whisk Animate, which turns images into animated clips.

Videos are provided in MP4 format at 720p resolution, with a digital watermark, SynthID, to ensure transparency. Google Labs is monitoring content generation for safety, incorporating red teaming, and allowing user feedback. The feature is rolling out to web and mobile users.

On another note, Google is also hiring a Research Scientist for "Post-AGI Research," highlighting ongoing advancements in AI. Read more.

Claude Gains New Research Tool, Voice Features, and Google Integration

Anthropic has launched several new features to expand its AI capabilities. The Claude Research tool enhances Claude’s ability to generate detailed responses by running multiple searches across internal and web sources. Coupled with a Google Workspace integration, this allows Claude to access Gmail, Calendar, and Docs for more personalized responses, with future updates promising broader content sources.

In addition, Anthropic has introduced a new voice mode for Claude, offering three distinct voices—Mellow, Airy, and Buttery—to rival similar features in ChatGPT. This comes alongside a strategic AWS partnership and a new premium tier for AI users, cementing Anthropic’s growing position in the competitive AI landscape. Read more.

• MIT introduces DISCIPL, a self-steering framework for constraint-aware LMs

• Cohere’s Embed 4 handles 200-page files, scanned docs, and handwriting

• Kuaishou Rolls Out Kling 2.0 with AI-Powered Multimodal Editor

• Hugging Face Acquires Pollen Robotics to Open Source Humanoid Robot

• Grok Studio Adds Code Execution and Google Drive Integration for Free Users

• Slopsquatting: One in five AI code snippets contains fake libraries

• Apple details how it plans to improve its AI models by privately analyzing user data

• Opera adds its Aria AI assistant to Opera Mini browser

• LMArena introduces Search Arena for evaluating search-augmented LLM systems

• What Engineers Should Know About AI Jobs in 2025

• 'Contagion' Writer Tests AI Storytelling in a Planned Sequel

• Figma sent a cease-and-desist letter to Lovable over the term 'Dev Mode'

• People are really bad at spotting AI-generated deep fake voices

• NVIDIA shifts AI supercomputer production to America in historic $500 billion move

• Huawei’s CloudMatrix 384 claims edge over NVIDIA’s GB300 in all core specs

• Meta AI will soon train on EU users’ data

• Project Maven reaches Europe: NATO selects Palantir's AI analytics system

• Gemini app rolling out Google Photos integration on Android

• Over-training large language models may make them harder to fine-tune

• Watch: Google is using AI and Pixel phones to understand what dolphins are saying

• AMD set to produce its CPU chip at TSMC plant in Arizona

• Machines could make medicine more human again

  

  5 new AI-powered tools from around the web

📄 InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

📄 PRIMA.CPP: Speeding Up 70B-Scale LLM Inference on Low-Resource Everyday Home Clusters 

📄 MineWorld: a Real-Time and Open-Source Interactive World Model on Minecraft

📄 xVerify: Efficient Answer Verifier for Reasoning Model Evaluations

📄 FUSION: Fully Integration of Vision-Language Representations for Deep Cross-Modal Understanding

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!