GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

Andrej Karpathy Joins Anthropic's Pretraining Team — OpenAI Co-Founder Returns to Frontier R&D

Andrej Karpathy announced on May 19 that he has joined Anthropic. He will work on the pretraining team under Nick Joseph — the phase of frontier model development where base capabilities are established through training on raw compute at scale.

“I think the next few years at the frontier of LLMs will be especially formative,” Karpathy wrote. “I am very excited to join the team here and get back to R&D.”

Anthropic confirmed Karpathy starts this week and will build a dedicated team focused on using Claude to accelerate pretraining research — putting the company’s production model in the loop of developing its own successors.

Why This Move Matters

Karpathy’s background runs through the foundational layer of the current AI stack. He co-founded OpenAI in 2015, where he worked on deep learning and computer vision before leaving for Tesla in 2017. At Tesla, he served as Director of AI and built the computer vision team behind Autopilot. He returned to OpenAI in 2022, then left again in 2023 to found Eureka Labs, an AI-native education platform. His neural networks from scratch course remains one of the most-cited introductory resources in the field.

His choice of Anthropic over OpenAI — where he is a founding member — is a signal. He made no public criticism of his former employer. But the move places one of the most recognized names in LLM development inside the team responsible for training Claude’s next generation of base models, at a moment when Anthropic is preparing to scale Claude Opus 4.7 successors under its $200B Google Cloud commitment and the SpaceX Colossus compute deal.

Pretraining Is Where It Starts

Pretraining is compute-heavy and has historically required the deepest concentration of ML expertise at any lab. It’s where base capabilities — reasoning, factual knowledge, code, long-context handling — are established before post-training alignment work begins.

Nick Joseph, who leads the pretraining team, has been a key figure in Claude’s architectural development. Adding Karpathy creates a pairing between one of the field’s most recognized researchers and the team making decisions about how Claude learns.

Karpathy said he remains passionate about education and plans to resume work on it eventually. For now, the focus is pre-training.

Pattern of Talent Movement

Karpathy is the second high-profile hire Anthropic has announced in May. Ross Nordeen, a founding member of xAI and former Tesla employee, announced his join earlier this month — on the same day Anthropic confirmed its compute deal with SpaceX’s Colossus 1 data center.

The pattern: talent from OpenAI’s founding cohort and from xAI moving to Anthropic, as the company commits to sustained compute access and prepares next-generation model releases. As of May 19, Anthropic sits at $900B in reported valuation negotiations, has $200B in Google Cloud compute committed, and is running agents at 97% performance against its own alignment researchers. Karpathy is arriving into a lab in an expansionary phase.