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

Here’s what we got for ya today:

Let’s get right into it!

Sign Up | Advertise | Tools

OpenAI's new GPT-5.5 isn't an update — it's the first complete rebuild of its base model since GPT-4.5. Everything that came out in between was post-training work on the same foundation. This time they started over.

The model scores 82.7% on Terminal-Bench 2.0, a test for complex command-line workflows that require planning and tool coordination, and 58.6% on SWE-Bench Pro, which grades real GitHub issue resolution end-to-end. It does this while matching its predecessor's speed and using fewer tokens to finish the same work. Engineers at NVIDIA who got early access put it bluntly: "Losing access to GPT-5.5 feels like I've had a limb amputated." OpenAI says 85% of its own staff now use Codex every week.

→ Read more

Not long ago the Trump administration labeled Anthropic a supply chain risk and the Pentagon took them to court. Now the White House is drafting an executive action to let federal agencies bypass that designation and access Mythos — Anthropic's most powerful model. Axios reported that one source described the effort as a move to "save face and bring em back in."

The core dispute is still unresolved. Anthropic refused to let the Pentagon use Claude for "all lawful purposes" — specifically drawing a hard line against mass domestic surveillance and fully autonomous weapons. OpenAI and Google both signed that agreement. Anthropic won't. But Mythos topped nearly every major security benchmark it was tested on, and the NSA is already using it. At some point the math on keeping it out of government stopped working.

→ Read more at Axios

The Zuckerberg-Chan Biohub is putting $500 million into a five-year effort to build AI that can model disease at the level of a single cell. The stated goal: cure all human disease. Biohub chief Alex Rives told Axios: "I think there is a path to building accurate predictive models of the cell." Of the $500M, $400M funds Biohub's own labs and $100M backs outside researchers through partners including Nvidia, the Allen Institute, the Human Cell Atlas, and the Human Protein Atlas.

The bottleneck is data. Most existing biological datasets cover roughly 1 billion cells. Rives says they need an order of magnitude more than that to make the models useful. The honest catch, per Rives: "We don't yet know what the slope of the scaling law is with cellular biology." They're funding the experiment to find out.

→ Read more

Meta's AI tools for businesses went from 1 million conversations a week at the start of 2026 to 10 million by late March. Zuckerberg disclosed the jump on the company's Q1 earnings call — the same call where Meta posted $56.3 billion in revenue, up 33% year over year. The tools run across Meta's messaging apps and are powered by Muse Spark, the first model out of Meta's Superintelligence Labs.

None of it costs businesses anything yet. Zuckerberg said on the call: "Business AIs today are currently free for most businesses on our messaging apps, but as we make more progress, we expect that we will also work towards establishing a longer-term monetization model." More than 8 million advertisers are already using at least one of Meta's AI ad tools, and those using its video generation feature are seeing conversion rates more than 3% higher in tests.

→ Read more