NeoLabsAn index of research-first AI labs
Add a lab
Home / Topics / Self-improving AI & automated research
Research area

Self-improving AI startups

Systems that run the experiments: AI that does AI research, recursive self-improvement. We track 14 labs in this area, which have raised $2.05B in disclosed rounds between them; the most valuable is Ineffable Intelligence, at $5.1B.

14labs
$2.05Bdisclosed raised
$11Breported valuations
LabTopicLocationFounders fromFoundedLast roundValuationRaisedStatus
Ineffable IntelligenceSuperlearner AI that learns from experience through reinforcement learning rather than hum…site Self-improving AI & automated research London, UK ex-Google DeepMind, ex-InstaDeep 2026 Seed $1.1BApr 27, 2026 · Sequoia, Lightspeed $5.1B $1.1B Research only
Recursive SuperintelligenceAI that automates model architecture, training, evaluation and research direction: recursi…site Self-improving AI & automated research San Francisco, US ex-Salesforce, ex-Meta, ex-Google DeepMind, ex-OpenAI, ex-Cresta 2026 Seed $650MMay 13, 2026 · GV (Google Ventures), Greycroft $4.65B $650M Research only
MirendilSelf-improving AI for scientific R&D: a system that builds systems, sold as a research pla…site · in Self-improving AI & automated research San Francisco, US ex-Anthropic, ex-xAI, ex-MIT 2026 Seed $200MJun 26, 2026 · Andreessen Horowitz, Kleiner Perkins $1B $200M Research only
Core AutomationThe world's most automated AI lab: Ceres, a continual-learning model that updates its weig…in Self-improving AI & automated research San Francisco, US ex-OpenAI, ex-Google DeepMind 2026 · $4Best. · Stealth
EngramA learned memory layer that needs 100x fewer tokens. Self-improving AI & automated research San Francisco, US ex-Stanford, ex-Cornell, ex-Columbia, ex-UC Berkeley 2026 · $600Mest. · Stealth
TrajectoryA continual-learning layer trained on agent trajectories.site · x Self-improving AI & automated research San Francisco, US ex-Windsurf, ex-Google DeepMind, ex-Stanford 2026 · $115Mest. · Stealth
Adaption LabsAdaptation beats scale: smaller models that keep learning on the fly, gradient-free contin…site Self-improving AI & automated research San Francisco, US ex-Cohere 2025 Seed $50MFeb 4, 2026 · Emergence Capital · $50M Research only
AutoscienceAutonomous AI scientists that automate the end-to-end research pipeline.site · x · in Self-improving AI & automated research San Mateo, US ex-Google DeepMind 2025 · · · Research only
CursiveAdaptive software and agents that improve from use. Self-improving AI & automated research London, UK ex-Google DeepMind, ex-NYU 2026 · · · Stealth
Discovery LoopAutomating the scientific method, from four of Google's most senior researchers. Self-improving AI & automated research Palo Alto, US ex-Google DeepMind 2026 · · · Stealth
InherentAI-native science: research agents that decide which scientific questions are worth asking…site · x · in Self-improving AI & automated research London, UK ex-Google DeepMind, ex-Microsoft 2026 Seed $50MMay 27, 2026 · Index Ventures, Radical Ventures · $50M Research only
NeoCognitionAgents that learn a world model of work, from the Mind2Web and SeeAct team. Self-improving AI & automated research Palo Alto, US ex-Ohio State, ex-Microsoft 2025 · · · Research only
Oak LabA trillion-parameter continual-learning agent on 20 watts, from the reinforcement-learning… Self-improving AI & automated research Toronto, CA ex-Google DeepMind, ex-Alberta 2026 · · · Stealth
PoetiqModel-agnostic self-improving reasoning; 'the fastest path to superintelligence via practi…site · x · in Self-improving AI & automated research Mountain View, US ex-Google DeepMind 2025 · · · Research only

News