GPT-4.1 Incoming

TODAY IN AI

1. DeepMind’s latest model could help solve AI’s biggest challenges: DeepMind has introduced AlphaEvolve, a coding agent designed to tackle complex technical issues—ranging from optimizing data center operations and refining chip architectures to enhancing the training processes of models like Gemini. When presented with over 50 publicly available math problems, AlphaEvolve successfully solved 75% of them, and in 20% of cases, it even outperformed the best-known solutions. The system was developed by merging Gemini’s generative creativity with a rigorous validation framework that evaluates and ranks each solution.

2. OpenAI finally brings GPT-4.1 to ChatGPT: OpenAI has announced that GPT-4.1 will be rolling out to Plus, Pro, and Team subscribers, while free-tier users will have access to a lighter “mini” variant. As part of this update, GPT-4o will be temporarily phased out, but there’s no need for concern: GPT-4.1 promises improvements across the board. It blends the advanced coding capabilities of previous “o” models with the responsiveness of OpenAI’s foundational architectures. Though both GPT-4.1 and its mini version were unveiled in April, they were previously limited to developer previews—now they’re opening up to the wider ChatGPT user base.

3. Anthropic is about to break its silence: Claude’s creator is on the verge of launching new iterations of its Sonnet and Opus models, which for the first time will integrate robust reasoning with seamless tool utilization. These upgrades will allow the AI not only to generate ideas but also to autonomously verify them by querying external databases and interacting with applications. A similar feedback loop will be applied to coding: the models will test intermediate results, correct any errors, and then present a polished final output. While the exact release date remains unannounced, these enhanced models are expected to arrive within the coming weeks.

AI & TECH NEWS
Everything else you need to know today

• 🔬 Molecular Map: Meta’s Fundamental AI Research team has published an open-source model and accompanying dataset designed to accelerate drug and materials discovery. Trained with an astonishing 6 billion compute hours—equivalent to running 100 million individual molecular simulations—the framework enables rapid prediction and design of novel compounds.

• 🔊 Good Vibrations: In a new collaboration, Stability AI and Arm have released an open-source audio generation model that’s lightweight enough to run entirely on a smartphone. It can produce short audio snippets—think drum loops, sound effects, or musical riffs—of around 11 seconds, all without relying on cloud services.

• 🎬 Scene Machine: Alibaba’s latest open-source innovation, Wan2.1-VACE, lets creators transform static images, text prompts, and video clips into fully fledged videos, then refine them using precise, natural-language instructions. (Click here to see a demo in action.)

• 💰 Funding: Legal-tech powerhouse Harvey is on track for a $250 million funding round that would boost its valuation from $3 billion to $5 billion. Meanwhile, data analytics leader Databricks has agreed to acquire Neon—an open-source database platform—for roughly $1 billion.

Thank you for reading!