What is General Reasoning?
General Reasoning (GR) is a London-based AI research lab dedicated to building the infrastructure for AI systems that can reason, plan, and act coherently over long horizons. Unlike current large language models that excel at procedural tasks but falter in dynamic, non-stationary environments, GR focuses on developing models that can adapt and coordinate reliably in real-world scenarios. The company's research spans reinforcement learning, multi-agent systems, and environment design, aiming to solve civilization-scale problems.
Key features include a benchmark for long-horizon sequential decision-making, which reveals that even frontier models from Google, OpenAI, Anthropic, and xAI struggle with tasks like predicting football season outcomes or profitable betting. GR also offers an open resource for community RL environments with simple API endpoints and autoscaled infrastructure, fostering collaboration and experimentation.
Benefits of GR's approach include improved reliability in complex, changing environments, enabling applications in finance, logistics, autonomous systems, and strategic planning. Use cases range from long-term investment strategies to multi-agent coordination in supply chains or disaster response.
Technically, GR builds on innovations from its team, who previously drove open language models at Meta, including Galactica, Llama 2, and Llama 3. They contributed thinking tokens, data-scarce training methods, and early RL techniques for LLMs that achieved state-of-the-art reasoning results. GR emphasizes rapid iteration ('Be Wrong Quickly'), continuous improvement ('Get Good'), collaborative culture ('Genius is a Team Effort'), and machine-centric design ('Build for Machines'). The team is mission-driven and seeks individuals passionate about advancing AI's ability to tackle long-horizon challenges.
Use Cases
Key Features
long horizon reasoning sequential decision making multi agent coordination reinforcement learning environment design open RL resources autoscaled infrastructureWho is it for?
AI researchers, reinforcement learning engineers, multi, agent system developers, finance professionals, logistics planners, autonomous systems engineers, strategic planners
Similar Tools
Details
AgentSwarms is a browser-based platform that teaches you to build and run agentic AI systems. With c...
free
Details
Crucible AI Braintrust is a revolutionary decision support AI designed to challenge your assumptions...
Free trial
Details
Fable Simulation is an innovative AI simulation game that puts you at the forefront of artificial in...
Freemium