Skip to main content

Core Automation

Category: Foundation Models / LLMs

AI research lab building continual-learning models and new architectures beyond transformer scaling, founded by ex-OpenAI VP of Research Jerry Tworek. Core Automation was founded in 2026. The company is led by Jerry Tworek. Based in San Francisco, USA. Team size: 11-50. Total funding raised: $100M. Latest round: Venture Round. Key investors include Nvidia, Spark Capital, Accel.

Founded
2026
Headquarters
San Francisco, USA
Team size
11-50
Total funding
$100M

Value proposition

Pursuing new learning algorithms and architectures that supersede large-scale pretraining/RL, aiming for models that continually learn from real-world experience with ~100x less training data, enabling small teams to do frontier-scale work.

Products and solutions

Flagship research project codenamed 'Ceres' — a continual-learning model architecture targeting a 100-fold reduction in training data versus current state-of-the-art models, broader mission to build 'the most automated AI lab in the world' starting by automating its own research process.

Unique value

Continual learning models that update from real-world/production experience post-training, plus architectures designed to scale better than transformers — a contrarian bet against the dominant scale-more-data-and-compute paradigm.

Target customer

Not yet commercialized — currently a pure research lab in fundraising/talent-acquisition stage; no disclosed enterprise customers.

Industries served

Frontier AI research / foundation models

Technology advantage

Founding team combines frontier model-building experience (o1/o3, Codex, GPT-4/5) with recruits poached from OpenAI, Anthropic, and Google DeepMind; small-team-plus-highly-capable-agents operating model designed to compound research velocity.

How they differentiate

Explicitly rejects the 'scale current recipe' approach (larger models, more data, static deployment) in favor of continual learning algorithms and non-transformer architectures, run by an extremely small, highly automated team rather than a large org.

Main competitors

OpenAI, Anthropic, Google DeepMind (frontier AI labs also competing for continual-learning/next-gen-architecture talent and capital)

Key partnerships

Reported investor/compute relationship with Nvidia (participated in initial funding round), no other formal partnerships publicly disclosed.

Major milestones

Founded ~January-March 2026 following Tworek's departure from OpenAI, closed an initial ~$100M round with Nvidia, Spark Capital, and Accel, recruited senior researchers from OpenAI, Anthropic, and Google DeepMind within weeks of launch, as of the anchor news (Aug 2026), reportedly in talks to raise an additional $300M-$1B at a $4-5B valuation, a ~4x valuation jump in weeks — this larger round appears to be in progress, not yet confirmed closed.

Market positioning

Ultra-early-stage but extremely high-profile frontier AI lab, positioned as a contrarian alternative research bet attracting outsized investor interest just weeks after founding.

Geographic focus

Global frontier AI race, primarily US-centered (competing with SF/Bay Area labs)

About Jerry Tworek

OpenAI VP of Research (2019-2026); led development of o1/o3 reasoning models and post-training of GPT-4/GPT-5; primary researcher behind Codex (foundation for GitHub Copilot); Uniwersytet Warszawski (University of Warsaw) education; started as quantitative trader turned deep RL researcher.

Latest news about Core Automation

More Foundation Models / LLMs companies

Official website: