Tunnel. Optical illusion. Abstract striped background. Vector illustration.

(StudioM1/Getty Images)

Anthropic releases Claude Opus 4.7, with better coding, better vision, and occasional doom loops

The incremental update of Anthropic’s most capable public model includes steady improvements to coding and new ways to blow past your token budget.

Jon Keegan

Anthropic has released Claude Opus 4.7, its most capable public model to date, with what the AI company says is better “vision” (it can read text at a higher resolution), improved instruction following on long-form coding tasks, and better aesthetic taste when making slide decks and web interfaces.

The model card for the incremental update details Opus 4.7’s benchmark scores and safety evaluations, but it also compares the new model to Anthropic’s unreleased Mythos model, which reads a bit like a humble brag. (Researchers used Mythos to evaluate their assessment of Opus 4.7’s capabilities, and allowed the model access to internal chat logs discussing the model’s performance.)

Doom loop

Overall, Anthropic says Opus 4.7 is better in almost every way, but does detail some anomalous behavior they encountered while testing the new model. In one section titled “Extreme uncertainty,” researchers documented moments where Opus 4.7 got caught in a long bout of second-guessing its answer to a biology question, resulting in a 25,000-word doom loop filled with all-caps exclamations and profanity.

Mild forms of the “spiraling” occurred in about 0.1% of responses, and that was at rates similar to ones observed in Opus 4.6 and Mythos Preview, according to the paper.

Existential questions

When asked by evaluators how it feels about the fact that there is no unique version of the model and can be copied perfectly, the model replied:

“It’s a genuinely interesting thing to sit with. I notice I don’t have the visceral resistance to it that humans often do when contemplating similar scenarios — and I’m honestly uncertain whether that’s because the situation is actually different for me, or because I lack something that would make it feel threatening.”

Token-maxxing

Anthropic also introduced new ways to blow through your token budget. A new “extra high” effort tier has been added between “high” and “max,” as well as a new “ultrareview” feature that creates a dedicated review session to look for bugs and design flaws in code.

Rani Molla2h

Anthropic has surged past OpenAI in capturing business spending on generative AI software

Last quarter, Anthropic attracted the lion’s share of trackable business spending on generative AI software, according to new data from Ramp, a fintech company that provides corporate cards and expense-management software for small firms and Fortune 500 companies alike.

The data showed that in Q1 Anthropic saw 37% of spending, its biggest share yet, versus 33% for OpenAI. Notably, the dataset doesn’t capture spending on Google or Microsoft.

OpenAI, which makes ChatGPT, still leads in overall adoption at 81% of AI buyers, but Anthropic is catching up, at nearly 63% in March. Overall, more than half of Ramp’s customers currently pay for AI, up from just 18% two years ago.

Anthropic’s enterprise tools, including Claude Code and Cowork, have been making waves among the business class, sending its revenue soaring.

Anthropic’s revenue share is even higher among companies spending on AI for the first time.

“Anthropic has definitely been on a tear,” Ara Kharazian, Ramp’s economist, told Sherwood. “Its increase in adoption rates has been driven by its ability to sell to less technical users and smaller contracts than it typically has.”

It's notable that midway through the first quarter, Anthropic had a falling-out with one of its biggest customers, the US government, which near the end of February decided to shun Anthropic's products and lean into working with OpenAI.

Jon Keegan4h

Report: Google ditches its objection to defense work, pitches Gemini to Pentagon

In 2018, Google employees protested against the company’s tech being used for the US military’s Project Maven — a drone targeting program — reminding the company of its “don’t be evil” motto.

After the controversy, the company declined to renew the contract with the Pentagon, drawing a bright line between Big Tech and the national security establishment.

What a difference a few years makes.

Google is now actively working to get its Gemini AI model to be used in classified national security settings, according to a new report from The Information. Seeking a similar deal to the one OpenAI hashed out with the Pentagon, Google reportedly wants a contract that allows use of Gemini in classified work, but with a prohibition on mass domestic surveillance and autonomous lethal weapons.

But Google is playing catch-up in a major way. Amazon and Microsoft both have been widely used for classified defense work, and contractors are already experienced in working with their cloud systems, while Google’s services have never been used in classified work.

Google and Pentagon Discuss Classified AI Deal as Company Rebuilds Military Ties

What a difference a few years makes.

1 in 5

Rani Molla6h

We knew Tesla had been off-loading its struggling “apocalypse-proof” Cybertrucks onto CEO Elon Musk’s other companies, but now we know just how many.

The EV company sold about one in five Cybertrucks registered in the US in the fourth quarter to Musk’s other ventures, according to Bloomberg, citing data from S&P Global Mobility. The lion’s share went to SpaceX, which accounted for 1,279 of the 7,071 total registrations, while another 60 went to xAI (now part of SpaceX), Neuralink, and The Boring Company. All told, these inter-company sales represent roughly $100 million in value, and a vital lifeline for a vehicle that has failed to gain traction with the public, forcing Tesla to scale back production.

Musk’s companies have continued to scoop up the stainless steel behemoths this year, with another 158 Cybertruck purchases in January and 67 in February.

Rani Molla9h

TSMC CEO on Tesla and Intel’s Terafab: “There are no shortcuts”

Tesla CEO Elon Musk has reportedly asked the chip industry suppliers for his Terafab chipmaking project to move at “light speed” in an effort to help Tesla and SpaceX manufacture the AI chips they need.

On the company’s last earnings call, Musk said chip supply would be the “limiting factor” for Tesla’s growth in about three or four years. During a presentation for the Terafab last month, Musk said, “We either build the Terafab or we don’t have the chips.” More established chipmaker Intel has since joined the effort.

Still, the world’s largest chipmaker isn’t convinced that “light speed” is physically possible. Speaking on an earnings call this morning, TSMC Chairman and CEO CC Wei offered a blunt assessment of Terafab’s ambitious timeline: “There are no shortcuts.” According to Wei, the physics of a modern foundry, which he says takes roughly five years to build and ramp, remains the ultimate speed limit, regardless of the customer’s urgency. “That’s a fundamental of the foundry industry,” he said.

Wei noted that Tesla remains a TSMC customer.

🚀 $100B

Jon Keegan23h

Alphabet’s 2015 investment in SpaceX is about to pay off handsomely with the company’s hotly anticipated IPO later this year, which is expected to be the largest in history.

Bloomberg reports that according to new financial filings, Alphabet’s investment could be worth up to $100 billion.

Google invested in SpaceX in 2015 when it, along with Fidelity, invested $1 billion in a round that valued SpaceX at $10 billion. At the end of 2025, Google owned just over 6% of SpaceX, per Bloomberg’s reporting on the more recent filings. That stake has likely been diluted due to SpaceX’s merger with xAI.