closeAI Research
Pioneering safe, beneficial artificial general intelligence.
Our research focuses on frontier reasoning architectures, multimodal understanding, and verifiable alignment systems that empower human potential.
GPT-5.6: The Next Paradigm in Multistep Deductive Inference
Delivering 3x faster chain-of-thought verification with 94.2% accuracy on competitive Olympiad mathematics and complex algorithmic engineering benchmarks.
Try it nowSelected Publications
Updated August 2026Frontier Reasoning with Hierarchical Self-Consistency Chains
We introduce a novel architecture for multi-turn deductive reasoning that evaluates candidate proof steps through parallel search trees.
closeAI Research Team
Multimodal Video Perception and Spatiotemporal Action Grounding
Exploring continuous high-frame-rate frame representations for complex spatial manipulation and dynamic camera reasoning.
Vision Intelligence Lab
Self-Verifiable Code Synthesis Across Polyglot Codebases
Achieving state-of-the-art benchmarks on SWE-bench through automated linting, test synthesis, and iterative sandbox execution.
Codex Foundations
Provable Alignment & Red Teaming for Autonomous Agents
Formulating formal safety bounds and guardrail boundaries for tool-augmented agents operating in high-stakes environments.
Safety & Alignment Group