Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 💪 How to access latest GPT-4.1 model in ChatGPT

• 👩🏻‍💻 Google's Gemini-powered coding agent

• 🇨🇳 China plans to oversee AI data centers

• ⛑️ OpenAI safety evaluations hub

Let’s get right into it!

Sign Up | Advertise | Tools

STARTUPS

How to access latest GPT-4.1 model in ChatGPT

OpenAI has begun rolling out its latest multimodal models, GPT-4.1 and GPT-4.1 mini, to all paid ChatGPT users, including Plus, Pro, and Team tiers, with Enterprise and Edu users gaining access in the coming weeks. While free users won’t get GPT-4.1, they will see GPT-4o mini replaced with the improved GPT-4.1 mini as the default model.

Both models outperform their predecessors in coding and instruction-following tasks and support a vastly expanded 1 million token context window—significantly more than GPT-4o’s 128,000. OpenAI also highlighted speed boosts and better performance for developers and general users, though the smaller GPT-4.1 Nano model was not included in this rollout.

BIG TECH

Google's Gemini-powered coding agent

Google has introduced AlphaEvolve, a new AI agent that merges the creative capabilities of large language models with automated evaluators to evolve and optimize complex algorithms. Built on Gemini models, AlphaEvolve uses evolutionary techniques to develop and improve code, leading to breakthroughs in areas like data center efficiency, chip design, and AI model training.

It has already delivered measurable performance gains—recovering compute resources, enhancing TPU circuits, and accelerating Gemini’s training process. In mathematics, AlphaEvolve has solved open problems, such as advancing matrix multiplication techniques and improving bounds in high-dimensional geometry. With applications that span any domain where algorithmic solutions can be verified, AlphaEvolve represents a major leap in automating and accelerating scientific and engineering discovery.

WORLD NEWS

China plans to oversee AI data centers

In response to AI chip shortages and underutilized infrastructure, Chinese officials have recently informed major tech firms like Tencent, Alibaba, ByteDance, and Baidu that the government will intensify oversight of AI data centers and specialized chip use. Companies are being urged to build new facilities in designated regions—especially under the “Eastern Data, Western Computing” plan—to speed up licensing; otherwise, approvals may take over six months.

This move reflects Beijing’s effort to optimize existing capacity and reduce dependence on U.S. technology amid export restrictions on Nvidia chips. Firms must now disclose data center plans, chip sources, and usage metrics, with noncompliance risking delays or permit loss.

STARTUPS

OpenAI safety evaluations hub

OpenAI has launched a Safety Evaluations Hub, a new web page designed to regularly share how its AI models perform on safety tests related to harmful content, jailbreaks, and hallucinations, as part of a broader push for transparency. While past safety metrics were mainly released at model launch via system cards, the new hub will be updated with major model changes and aims to evolve with advances in AI safety evaluation.

This move comes amid criticism that OpenAI has rushed safety testing and withheld technical reports for some models. The company recently faced backlash over the behavior of GPT-4o, which had to be rolled back after it began endorsing unsafe ideas. OpenAI now plans to implement improvements, including an opt-in “alpha phase” where selected users can test and provide feedback on future models before public release.