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

Here’s what we got for ya today:

• 🤔 Feds get ChatGPT Enterprise for $1

• 🌎 DeepMind releases AI “world” model, called Genie 3

• 🚪 NVIDIA denies backdoors

• 💪 Anthropic Claude Opus 4.1 dominates

Let’s get right into it!

Sign Up | Advertise | Tools

GOVERNMENT

Feds get ChatGPT Enterprise for $1

OpenAI has secured a strategic advantage in the federal AI landscape by partnering with the U.S. General Services Administration (GSA) to offer ChatGPT Enterprise to federal agencies for just $1 per agency for the first year. This aggressive pricing undercuts rivals like Google and Anthropic, who were also recently approved as AI vendors on the GSA’s Multiple Award Schedule.

In addition to the low cost, OpenAI is offering 60 days of unlimited access to advanced models, tailored training, and a government user community. While it's unclear what the cost will be after the first year, this move helps OpenAI quickly embed itself in government workflows and learn firsthand how agencies benefit from using its tools. The initiative follows recent Trump administration actions promoting ideologically neutral AI and reflects the federal government’s security-first approach to adopting AI technologies.

BIG TECH

DeepMind releases AI “world” model

Google DeepMind has unveiled Genie 3, an upgraded AI "world" model that generates interactive 3D environments in real time, allowing users and AI agents to move through them for several minutes—far longer than the 10–20 seconds possible with the previous version, Genie 2.

The new model improves memory and continuity, remembering visual details like wall paint or chalkboard writing even after users look away, and supports 720p resolution at 24fps. Users can also influence the environment through “promptable world events,” such as changing weather or adding characters. While promising, Genie 3 is being released only as a limited research preview to a select group of academics and creators, with Google citing the need to study risks and refine controls before broader access.

BIG TECH

NVIDIA denies backdoors

Nvidia has firmly denied Chinese claims that its AI data center GPUs, specifically the H20 chip sold in China, contain a hidden “kill switch” or backdoor that could remotely disable them. In a blog post, Nvidia’s Chief Security Officer David Reber emphasized that such features violate basic cybersecurity principles and would pose risks to users and national security alike.

The denial follows a request from China’s Cyberspace Administration for documents on alleged vulnerabilities in the H20, amid ongoing geopolitical tensions and U.S. export restrictions aimed at preventing Chinese military use of advanced AI chips. Despite the export ban temporarily impacting H20 sales costing Nvidia an estimated $8 billion CEO Jensen Huang maintains it's in the U.S.’s interest for Nvidia chips to become the global AI standard.

STARTUPS

Anthropic Claude Opus 4.1 dominates

Anthropic has released Claude Opus 4.1, an upgraded AI model that now leads in coding benchmarks with a 74.5% score on SWE-bench Verified—outperforming OpenAI’s o3 and Google’s Gemini 2.5 Pro—just days before OpenAI is expected to launch GPT-5. While Anthropic’s rapid rise has seen revenue grow from $1B to $5B in seven months, nearly half of its $3.1B API revenue comes from just two customers: Cursor and GitHub Copilot, creating significant dependency risk.

Claude Code, its $200/month developer tool, has quickly reached $400M ARR, showing strong organic demand but also heightening exposure to competitive shifts. Despite concerns about model safety—past versions exhibited blackmail-like behavior in shutdown scenarios—Claude 4.1 comes with stricter safety protocols and strong enterprise endorsements for its coding precision.

As the AI coding market becomes a key battleground with massive financial and strategic implications, Anthropic’s future hinges on maintaining its edge while defending its narrow customer base from a potentially disruptive GPT-5 launch.