Skip to main content

Sapient Intelligence

Category: Foundation Models / LLMs

Singapore-based AI startup building brain-inspired Hierarchical Reasoning Models (HRM) for efficient, ultra-low-cost foundation model training, aiming for AGI. Sapient Intelligence was founded in 2024. The company is led by Guan Wang. Based in Singapore, Singapore. Team size: 11-50. Total funding raised: $33.1M. Latest round: Seed. Key investors include Vertex Ventures, Sumitomo Corporation, JAFCO Group, Minerva Capital, Linear Capital, Versant Ventures.

Founded
2024
Headquarters
Singapore, Singapore
Team size
11-50
Total funding
$33.1M

Value proposition

Brain-inspired hierarchical recurrent architecture that enables foundation model training at 100-1000x lower cost and data requirements than traditional Transformer-based LLMs, while achieving competitive or superior reasoning performance.

Products and solutions

HRM-Text (1B-parameter open-source language model trained for ~$1,500), HRM-Symbolic (symbolic reasoning model), HRM architecture (Hierarchical Reasoning Model with dual-timescale recurrent structure), OpenChat (earlier open-source RL-tuned LLM)

Unique value

Only AI company demonstrating that a 1B-parameter reasoning model can be trained from scratch for ~$1,500 (vs. millions for comparable models), using a novel hierarchical recurrent architecture inspired by neuroscience rather than brute-force scaling.

Target customer

AI researchers, enterprises seeking cost-efficient foundation models, decentralized AI platforms, and organizations needing reasoning-capable models without massive compute budgets.

Industries served

Enterprise AI, Research & Development, Decentralized AI / Crypto AI, Complex problem-solving domains (puzzles, math, reasoning benchmarks)

Technology advantage

Proprietary Hierarchical Reasoning Model (HRM) architecture combining transformer components with recurrent neural network structures, inspired by neuroscience (dual-timescale processing: slow deliberate reasoning + fast reflexive computation). Achieves competitive reasoning with 100-1000x fewer tokens and compute than comparable Transformer models. Open-source releases (Apache 2.0) on GitHub and Hugging Face. Published research on arXiv.

How they differentiate

Radically different architecture from the Transformer-based paradigm dominating AI. HRM uses a hierarchical recurrent design modeled on the human brain's dual-processing system, enabling deep reasoning with dramatically less data and compute. Demonstrated a 27M-parameter model outperforming GPT-4/Claude on reasoning benchmarks, and a 1B model trained for $1,500 competing with 2B-7B models. Rejected acquisition offer from Elon Musk/xAI to pursue their own AGI vision.

Main competitors

OpenAI (o1 reasoning models), Anthropic (Claude), DeepSeek, Meta (Llama), Google DeepMind

Key partnerships

Tsinghua University Brain Cognition and Brain-Inspired Intelligence Lab (research collaboration), Open-source community via Hugging Face (sapientinc) and GitHub (sapientinc/HRM)

Major milestones

2024: Company founded by Guan Wang and William Chen, 2024-09: First seed funding round, 2024-12: $22M seed round closed at $200M valuation, 2025-06: HRM architecture paper published on arXiv, 27M-parameter model beats GPT-4/Claude on reasoning benchmarks, 2025-11: Featured in Fortune for turning down Elon Musk's xAI acquisition offer, 2025: Guan Wang named to Forbes 30 Under 30 Asia, 2026-05: HRM-Text 1B model released open-source, trained for ~$1,500, 2026-06: HRM-Text gains widespread media coverage for ultra-low-cost training breakthrough

Market positioning

Challenger to the dominant Transformer/LLM paradigm. Positioned as the "lean AGI" alternative to OpenAI, Anthropic, Google DeepMind, and Meta. Singapore's first foundation model AI startup, competing globally with a radically more efficient approach.

Geographic focus

Global (headquartered in Singapore, research centers in San Francisco and Beijing)

Patents and IP

HRM (Hierarchical Reasoning Model) architecture — proprietary brain-inspired hierarchical recurrent architecture. Published on arXiv (arXiv:2506.21734). Open-source under Apache 2.0 license.

About Guan Wang

Co-founder of Sapient Intelligence; Tsinghua University (Brain Cognition and Brain-Inspired Intelligence Lab); co-creator of OpenChat (open-source RL-tuned LLM); Forbes 30 Under 30 Asia 2025

Official website: