Core Automation
Core Automation is a San Francisco-based AI research lab, founded in 2026 by former OpenAI VP Jerry Tworek, developing continual learning systems and non-Transformer architectures. The pre-revenue lab is building automated research tools and pursuing a $5B+ valuation.
- Company typePrivate
- Founded2026
- HeadquartersSan Francisco, United States
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What Core Automation does
Core Automation is a privately held AI research lab founded in January 2026 in San Francisco by Jerry Tworek, a former OpenAI research vice president who led development of the o1, o3, and GPT-5 model lines. The company's stated objective is to build 'the world's most automated AI lab' by developing AI systems that optimize and automate work, beginning with the lab's own research processes. Core Automation is pursuing novel learning algorithms and architectures designed to supersede large-scale pretraining and reinforcement learning, with its flagship project Ceres targeting continual learning capabilities that reduce training data requirements to 1% of current state-of-the-art while exploring non-Transformer architectures. Supporting technical work includes KernelBot (automated GPU kernel generation), KernelGuard (reward hacking prevention in code evaluation, to be presented at ICML), Project Popcorn (PyTorch-to-Triton synthetic data generation), and the GPU MODE community platform, which has collected over 500,000 kernel submissions from researchers worldwide.
The company operates with a small team (11-50 employees) of researchers recruited from frontier AI labs, including Rohan Anil (ex-Google DeepMind), Anmol Gulati (ex-Anthropic), and co-founder Mark Saroufim (ex-Graphcore, co-founder of GPU MODE). Core Automation's bet is that the next wave of frontier research will come from small teams augmented by highly capable AI agents rather than from scaling existing paradigms. As of the input data, the company is pre-revenue with no commercial products or pricing announced, and is seeking $500 million to $1 billion in funding at a valuation exceeding $5 billion, with $100 million already raised in April 2026 from Accel, NVIDIA, and Spark Capital.
Core Automation firmographics
Firmographics- Name
- Core Automation
- Legal name
- Core Automation
- Website
- https://coreauto.com
- Company type
- Private
- Founded year
- 2026
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Core Automation is a San Francisco-based AI research lab, founded in 2026 by former OpenAI VP Jerry Tworek, developing continual learning systems and non-Transformer architectures. The pre-revenue lab is building automated research tools and pursuing a $5B+ valuation.
- Ownership category
- akta.pro rank
Core Automation industry classification
Industry- Product category
- AI Research and Frontier Model Development
- NAICS
- Computer Systems Design and Related Services (54151)
- SIC
- Services-Prepackaged Software (7372)
- akta.pro primary industry
- AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI)
Keywords
Where Core Automation is headquartered
LocationHeadquarters
- HQ city
- San Francisco
- HQ country
- United States
- HQ region
- North America
Markets served
Core Automation business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Technology or R&D, Personnel, Infrastructure, Operations
Revenue model
- Pre-revenue research lab: Core Automation is in early-stage development seeking funding. The company has no revenue or proven product at industrial scale. The business model for eventual commercialization of its AI research and continual learning systems has not been disclosed.
Go-to-market motion1 record
Marketing channels3 records
Core Automation product offering
Product offeringCore offering
Core Automation is an AI research lab developing new learning algorithms and architectures that supersede large-scale pretraining and reinforcement learning. Its flagship Ceres project pursues a continual learning model that requires 100x less training data than current state-of-the-art systems, while supporting products (KernelBot, KernelGuard, Project Popcorn) automate GPU kernel generation, evaluation, and anti-cheating for AI-driven systems research.
Product overview
Core Automation is an AI research lab developing a portfolio of products centered on automated AI systems that optimize and automate work, starting with research itself. The flagship Ceres model pursues continual learning with architectures beyond Transformers. Supporting products include KernelBot (automated GPU kernel generation), KernelGuard (reward hacking prevention), Project Popcorn (kernel generation training), and GPU MODE (kernel competition platform). These products form a self-improving ecosystem where AI systems generate kernels, evaluate submissions, prevent cheating, and continuously learn from the process.
Differentiator
Problem solved
Functional benefit
Products and services
- Ceres Core Automation's flagship continual learning AI model designed to reduce training data requirements to 1% of current state-of-the-art while exploring architectures beyond the Transformer paradigm, enabling learning in production environments. Target customers are organizations seeking frontier AI model capabilities with dramatically improved data efficiency.
- KernelBot AI system for automated GPU kernel generation that competes against human developers in kernel writing competitions and learns to produce competitive kernel implementations through CLI submissions. Target users are GPU programmers, ML researchers, and competition participants.
- KernelGuard AI-powered system designed to prevent reward hacking in automated code evaluation, using rules-based regex detection trained on human audit datasets to catch cheating submissions in kernel competitions. Target users are competition organizers and automated evaluation pipeline operators.
- GPU MODE Community platform and kernel competition website hosting GPU programming challenges with prize pools up to $1M, enabling Core Automation to collect high-quality kernel submission data and benchmark AI kernel generation systems. Target users are GPU programmers, ML researchers, and AI competition participants.
Quantifiable outcome
- 100× less training data required compared to current AI giants
- +1 more outcomes
Companies that use Core Automation
Customer profileIdeal customer profiles1 record
Core Automation technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability7 records
Feature4 records
Core Automation partnerships and signals
Strategic signalScale indicators6 records
Recent moves6 records
Expansion highlights5 records
Core Automation competitors and assessment
Company assessmentDirect peers
- Sakana AI: Direct peer as a research-driven AI lab explicitly focused on novel, non-transformer-dominant architectures and evolutionary/model-merging approaches. Closest peer in research philosophy to Core Automation's post-transformer thesis.
- OpenAI: Direct peer as a frontier AI research lab pursuing novel architectures and large-scale training. Core Automation's founder Jerry Tworek spent seven years as Research VP at OpenAI leading o1, o3, and GPT-5; Joanne Jang also co-founded there. The two compete for the same researcher talent and pursue overlapping research agendas.
- Google DeepMind: Direct peer as a frontier AI research organization. Core Automation hired Rohan Anil from Google DeepMind, and both pursue continual learning, novel architectures, and AI-for-science applications at the research frontier.
- Mistral AI: Direct peer as an emerging frontier AI lab building foundation models with significant capital and a research-oriented engineering culture. Competes for similar European/international research talent.
- Anthropic: Direct peer as a frontier AI safety/research lab. Core Automation recruited Anmol Gulati directly from Anthropic, and both pursue novel architectures and agentic AI capabilities with comparable research depth.
- xAI: Direct peer as an emerging frontier AI lab founded by ex-OpenAI leadership, pursuing novel model architectures and large-scale training with comparable funding scale and similar talent-poaching dynamics.
- Safe Superintelligence (SSI): Direct peer as a research-only frontier AI lab with no commercial product, founded by former OpenAI leadership, raising at multi-billion valuations. Operates with a similar small-team, deep-research model.
Emerging players
- Hugging Face: Emerging player as the de facto AI platform hosting datasets (KernelBook is hosted there) and tooling for open model development. Strategic platform partner and potential distribution channel for Core Automation's kernel/dataset assets.
- Modular AI: Emerging player focused on AI compiler, runtime, and kernel optimization through the Mojo language and MAX platform. Strong technical overlap with Core Automation's KernelBot, Project Popcorn, and PyTorch-to-Triton translation work.
- Together AI: Emerging player building AI compute infrastructure including GPU kernel optimization and inference runtimes. Operates in adjacent infrastructure territory to Core Automation's compiler/runtime work.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat4 records
Key risks5 records
Key highlights5 records
Customer concentration
Core Automation social profiles
Digital presenceCore Automation financial estimates
Financial estimateRevenue estimate
Valuation estimate
Core Automation leadership team
Management profileNumber of profiles
Profiles6 records
Core Automation funding detail
Funding detailFunding overview
Funding rounds2 records
Investors5 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Core Automation M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Core Automation
What does Core Automation do?
Core Automation is an AI research lab developing new learning algorithms and architectures that supersede large-scale pretraining and reinforcement learning. Its flagship Ceres project pursues a continual learning model that requires 100x less training data than current state-of-the-art systems, while supporting products (KernelBot, KernelGuard, Project Popcorn) automate GPU kernel generation, evaluation, and anti-cheating for AI-driven systems research.
Is Core Automation a public or private company?
Core Automation is a private company. It is classified as venture growth investor backed and is currently operating.
When was Core Automation founded?
Core Automation was founded in 2026. It employs 11 to 50 people.
Where is Core Automation based?
Core Automation is headquartered in San Francisco, United States, in the North America region.
How does Core Automation make money?
One revenue line is on record: pre-revenue research lab.
Who are Core Automation's main competitors?
Direct peers on record are Sakana AI, OpenAI, Google DeepMind, Mistral AI, Anthropic, xAI and Safe Superintelligence (SSI). Emerging players are Hugging Face, Modular AI and Together AI.
Does Core Automation have an API?
No public API is recorded for Core Automation.
What industry is Core Automation in?
Core Automation's product category is AI Research and Frontier Model Development. Its primary akta.pro industry code is HDAAAAAI, AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers). Its NAICS code is 54151 and its SIC code is 7372.