Karpathy Joins Anthropic to Work on Model Pre-Training

*Andrej Karpathy, a founder of OpenAI, moves to Anthropic to help run the large-scale training that defines Claude’s base capabilities.*

Andrej Karpathy has joined Anthropic’s pre-training team. The move places a founding member of OpenAI at the company behind Claude, in a role tied directly to the most compute-intensive stage of frontier model development.

Pre-training covers the large-scale runs that supply a model with its core knowledge and abilities. Anthropic has described this phase as both essential and one of the costliest parts of building its systems.

Karpathy previously served as Director of AI at Tesla and was among the original team at OpenAI. His new position focuses on the same pre-training work that produces the weights later refined through post-training and alignment steps.

No public statements from either company accompanied the announcement. The two source reports agree on the basic facts of the hire and the focus on pre-training, with no conflicting details.

The shift puts direct experience from OpenAI’s early model work inside Anthropic’s training infrastructure at a time when every major lab is racing to increase compute budgets for the next generation of base models. Pre-training remains the step that determines how much raw capability later stages have to work with, and adding someone who has already shipped production-scale runs at two other organizations changes the distribution of that specific expertise.

---

Sources:

• TechCrunch
• Neowin

{
  "excerpt": "OpenAI co-founder Andrej Karpathy joins Anthropic to work on the compute-heavy pre-training runs that build Claude’s core capabilities.",
  "suggestedSection": "ai",
  "suggestedTags": ["andrej-karpathy", "anthropic", "openai"],
  "imagePrompt": "Abstract forms of layered data streams converging into a dense central core inside a dim server hall. Muted color palette, cinematic lighting, 16:9."
}