Top AI researchers continue to jump between leading AI labs as competition for elite talent intensifies.

Why it matters: Hiring elite AI researchers is hard. Keeping them — even after offering enormous pay, prestigious titles and vast computing resources — is proving almost as difficult.

Between the lines: The churn reflects an unusual moment in which a small group of researchers can command extraordinary compensation while choosing among companies with different cultures, missions and technical resources.

Zoom out: Frontier AI now resembles a single, tightly connected ecosystem rather than a collection of isolated rivals.

• Companies battle fiercely for talent and customers while simultaneously investing in one another, buying one another's services, relying on the same cloud providers and hiring from the same small pool of researchers.

Driving the news: The latest high-profile move is Lilian Weng, co-founder of Mira Murati's Thinking Machines Lab.

• She announced last week she was leaving Thinking Machines, saying that being a co-founder was taking a toll on her health and she wanted a more focused position.
• Just days later, The Information reported that she was rejoining OpenAI to work on recursive self-improvement, the idea that AI systems could play a growing role in improving future generations of AI.
• Weng is the fourth Thinking Machines co-founder to leave within the past year, suggesting that even well-funded startups can struggle to retain their founding talent.

The big picture: Thinking Machines isn't the only lab struggling with retention.

• Google lost a pair of prominent researchers to rivals in June as Noam Shazeer left for OpenAI and Nobel Prize in Chemistry winner John Jumper went to Anthropic.
• Meta has spent heavily to lure leading researchers to Alexandr Wang's superintelligence operation, only to see several of those prized recruits quickly decamp, some for OpenAI.
• Last year, former OpenAI co-founder and former chief scientist Ilya Sutskever lost the CEO of his startup Safe Superintelligence to Meta.

Money matters. Some researchers receive offers that are difficult to refuse, while others want equity in companies like Anthropic or OpenAI before a potential IPO.

• A source familiar told Axios that Anthropic CEO Dario Amodei has expressed concern about new talent coming to the firm for the money rather than the mission.
• Some researchers also see a narrowing window in which their expertise commands a premium. They worry that their bargaining power could diminish as AI systems eventually automate more of the work to improve and develop new models.

Yes, but: Compensation alone doesn't explain the movement.

• Top researchers also say they care deeply about access to computing power, influence over what gets built and the freedom to pursue their preferred technical approach.
• And in a field where many participants believe their work could reshape the global economy — or determine which company develops transformative AI first — status, ambition and ideological differences can matter as much as money.

What they're saying: It's partly money and "some ego about changing the world," an executive tech recruiter told Axios.

• "There is definitely a desire to make sure that everyone secures enough financial security for themselves," Tara Shulman, a financial advisor who works with many newly minted tech millionaires, told Axios.
• Engineers who have had lucrative careers are asking themselves, "What if that changes ... it comes up very often, more so than you would think, with folks that have a ton of equity comp and are in a very successful financial position," she added.

Reality check: Constant movement can impose costs on the labs as well as the researchers.

• Research programs depend on trust, institutional knowledge and teams that work together over long periods. Frequent departures can interrupt projects, unsettle other employees and leave companies paying repeatedly to recruit or retain the same small circle of people.
• More than 400 former Apple employees now work at OpenAI. Apple is suing OpenAI, alleging the company used internal Apple codenames to extract confidential information from prospective hires.

The bottom line: AI companies are finding they can buy someone's time, but securing lasting allegiance is proving much harder.

Chinese AI lab DeepSeek released a powerful new coding model last week that charges pennies for vast amounts of code — the latest sign that some of the smartest software on Earth is rapidly becoming a commodity.

Why it matters: Tech giants are pouring hundreds of billions of dollars into the computing infrastructure powering the AI revolution. Yet the intelligence that infrastructure produces is getting cheaper by the week.

Zoom in: DeepSeek is the same Chinese startup that ignited a market meltdown last January by showing it could build a world-class AI model with far fewer resources than its U.S. rivals.

• Its newest model, V4 Flash, performs close to the level of Anthropic's Claude Opus 4.8, one of the industry's most capable systems, on tests of complex coding and autonomous software tasks.
• On Arena.ai's crowdsourced leaderboard for front-end coding, V4 Flash debuted ahead of Opus 4.8 — while delivering the best performance for its price of any model in its class.
• The price gap is staggering: DeepSeek charges about 28 cents for the same amount of output that costs $25 on Opus 4.8 — a 99% discount.

Zoom out: With Chinese models like Kimi K3 bearing down on the U.S. market, July ushered in a full-scale price war across the AI landscape.

• OpenAI slashed the price of GPT-5.6 Luna — its fastest, cheapest model for high-volume tasks — by 80% on Thursday, only three weeks after its launch.
• Google released three new Gemini "flash" models all focused on efficiency.
• SpaceXAI released Grok 4.5, Elon Musk's most capable model yet for coding, research and autonomous tasks, at the same price OpenAI originally charged for Luna before last week's cut.
• Meta quietly reversed course on its longtime embrace of open weights with Muse Spark 1.1, a closed-source model priced aggressively for developers.

The other side: Anthropic remains the clearest holdout, keeping its top-tier Claude models at premium pricing and betting that developers will pay extra for safety and precision.

Keep reading.

• OpenAI showed off "Astra" in D.C. last week. It's a new model family that can complete long-running tasks with multiple agents working together to solve difficult problems. (The Information)
• Apple capped bug reports from outside researchers due to an influx of AI slop reports that hallucinate security risks. (Financial Times)
• Google pulled a Google Earth feature that made creating deepfaked satellite imagery too easy. The company said it will develop additional safeguards before relaunching. (NPR)
• Snapchat said it's tweaking its Spotlight recommendation engine to avoid promoting AI-generated content. (TechCrunch)

This is a fascinating idea for a way to transfer files between two devices without needing any form of networking.