AGI and superintelligence labs
These labs say, in their own published words, that they are building artificial general intelligence or superintelligence. The list does not judge who is closest. It records who has made the claim, who founded the lab, and how much money has been confirmed behind it.
| Stated goal | HQ | Founders | ||||
|---|---|---|---|---|---|---|
| MMiniMaxGraduated | Develops multimodal foundation models and AI-native consumer and API products | 2021 | CN | $1.56B | $4.24B | Yan Junjie, Yun Yeyi |
| IIneffable IntelligenceCore | Reinforcement-learning research lab building a 'superlearner' that learns from experience, not human data ineffable.ai Ineffable Intelligence is an AI research company whose mission is to make first contact with superintelligence.checked 9 Nov 2026 | 2025 | London, GB | $1.1B | $5.1B | David Silver |
| SSafe SuperintelligenceCore | Research lab building a safe superintelligence as its single product | 2024 | Palo Alto, CA, US | $1B | $5B$32B reported | Ilya Sutskever, Daniel Levy, Daniel Gross |
| RRecursive SuperintelligenceCore | Lab pursuing recursive self-improvement: automated AI research (Eureka Machine); first model Flint recursive.com Our goal at Recursive is to build recursively self-improving superintelligence to automate knowledge discovery.checked 9 Nov 2026 | 2026 | London, GB | $650M | $4.65B | Richard Socher, Tim Rocktäschel, Yuandong Tian +5 |
| GGeneralistCore | Embodied foundation models for robots, starting with dexterity generalistai.com Generalist is a frontier AI research and product driven company building general intelligence for the physical world.checked 9 Nov 2026 | 2024 | San Francisco, CA, US | $600M | $2B$3B reported | Pete Florence, Andy Zeng, Andrew Barry |
| LLila SciencesCore | Builds AI models and automated laboratories for autonomous scientific research ('scientific superintelligence') lila.ai Building Scientific Superintelligence to solve humankind's greatest challengeschecked 9 Nov 2026 | 2023 | Cambridge, MA, US | $550M | $1.3B$8.5B reported | Geoffrey von Maltzahn, Molly Gibson |
| MMagicCore | Frontier code models to automate software engineering and AI research | 2022 | San Francisco, CA, US | $348M | — | Eric Steinberger, Sebastian De Ro |
| ZZhipu AIGraduated | Develops GLM large language models and related agent and API products | 2019 | Beijing, CN | $342M | — | Zhang Peng |
| SStandard IntelligenceEmerging | Trains computer-action models on video pretraining to build agents that explore and learn like humans | 2024 | San Francisco, CA, US | $75M | — | Galen Mead, Devansh Pandey |
| DDeep CogitoEmerging | AI startup building open-weight language models through post-training and iterative self-improvement deepcogito.com Deep Cogito is an AI startup based in San Francisco, dedicated to developing open source general superintelligence.checked 10 Nov 2026 | 2024 | San Francisco, CA, US | $43M | — | Drishan Arora, Dhruv Malhotra |
| DDeepSeekGraduated | AI research lab building open-weight large language models, with a commercial API huggingface.co DeepSeek (深度求索), founded in 2023, is a Chinese company dedicated to making AGI a reality.checked 9 Nov 2026 | 2023 | Hangzhou, Zhejiang, CN | — | —$51.5B reported | Liang Wenfeng |
| MMath Inc.Emerging | Builds AI agents that formalize mathematics and software into machine-checkable proofs in Lean | 2025 | San Francisco, CA, US | — | — | Jesse Han, Christian Szegedy |
| MModelBestGraduated | On-device (edge) large models: the open MiniCPM series and applications in phones, cars and robots modelbest.cn ModelBest is an AI large model technology innovation and application company, dedicated to creating safe and accessible artificial general intelligence.checked 15 Nov 2026 | 2022 | Beijing, CN | — | —$2.95B reported | Liu Zhiyuan, Li Dahai, Zeng Guoyang |
| PPerceptronEmerging | Multimodal vision-language models for physical AI (perception, reasoning and robot control) | 2024 | Bellevue, WA, US | — | — | Armen Aghajanyan, Akshat Shrivastava |
| ZZyphraCore | Open models, multisilicon (AMD-first) inference cloud and an agent product zyphra.com Zyphra is building the full stack for multisilicon open superintelligence.checked 9 Nov 2026 | 2021 | San Francisco, CA, US | — | —$5B reported | Krithik Puthalath, Beren Millidge |
Raised = sum of rounds with Confirmed status. Valuation = latest confirmed figure; a reported figure from talks is shown below it and is not confirmed. Sources for every figure are on each lab's page.
Latest funding rounds in this field
| Date | Round | Amount | Valuation | Lead investors | Status |
|---|---|---|---|---|---|
| 22 Oct 2026 | DeepSeek Funding round | $15B | — | not disclosed | Reported |
| 26 Aug 2026 | Perceptron Funding round | undisclosed | — | not disclosed | Reported |
| 26 Aug 2026 | Deep Cogito Series A | $43M | — | TQ Ventures | Confirmed |
| Aug 2026 | Generalist Series B extension | $200M | $3B | 8VC | Confirmed |
| 27 Jul 2026 | Safe Superintelligence Strategic | $5B | — | not disclosed | Reported |
| 15 Jul 2026 | ModelBest Funding round | undisclosed | $2.95B | not disclosed | Reported |
Where these labs get compute
Compute deals on the ledger in which a lab from this list is a party.
| Date | Buyer | Provider | Type | Value | Capacity | Status |
|---|---|---|---|---|---|---|
| 9 Apr 2025 | Safe Superintelligence | Google Cloud | Capacity contract | undisclosed | — | Confirmed |
| 29 Aug 2024 | Magic | Google Cloud | Capacity contract | undisclosed | — | Confirmed |
How a lab gets on this list
A lab is listed when its own site, launch post or filing names AGI, general intelligence or superintelligence as what it is building. The sentence is quoted in the table. A lab that is described that way only by the press is not listed.
Every lab here also meets the rules for a neolab profile: a research-led company with named founders, a stated focus, a headquarters and a funding status, each with a public source. The tier under each name follows the methodology.
What is not here. OpenAI, Google DeepMind, Anthropic, Meta and xAI state the same goal. They are frontier labs, not neolabs, and have no profile here; they appear on the ledger as buyers in compute deals.
Other fields
- World model labs: 8 labs
- Robotics and physical AI labs: 16 labs
- AI for science labs: 17 labs
A lab is missing, or its description is out of date? Suggest an entry with a public source.