AI Policy Wiki
Dashboard

Sapient Intelligence

medium confidence · updated 2026-06-06

Singapore-based AI research lab pursuing 'lean' general intelligence through brain-inspired model architectures. Builder of the Hierarchical Reasoning Model (HRM) — a 27M-parameter reasoning model — and, in May 2026, the 1.15B-parameter open model HRM-Text.

Sapient Intelligence is a Singapore-based AI research lab co-founded in 2024 by Guan Wang (CEO) and William Chen. Its stated mission is "lean general intelligence" — pursuing artificial general intelligence through small, compute-efficient, brain-inspired model architectures rather than through frontier-scale pretraining. The company is the developer of the Hierarchical Reasoning Model (HRM), a 27-million-parameter reasoning model, and of HRM-Text, a 1.15-billion-parameter open language model released in May 2026.

Founded2024
HeadquartersSingapore
FoundersGuan Wang (CEO), William Chen
TypeAI research lab (frontier-lab)
Notable modelsHRM (27M params), HRM-Text (1.15B params)

Overview

Where OpenAI, Anthropic, and Google DeepMind scale parameters and compute, Sapient argues that architecture, rather than scale, is the path to reasoning, positioning it against the scaling consensus. Founder Guan Wang frames the work in AGI terms — "human-level intelligence, and eventually beyond." Both of Sapient's models are released with open weights, placing the company in the open-weight camp at a time when several developers (notably Alibaba's Qwen3.7-Max) are moving proprietary.

Wang and Chen previously created and open-sourced OpenChat, an early open chat model. Fortune reported in November 2025 that the two founders turned down funding from Elon Musk to build the company independently (Source: fortune.com). Public reporting on Sapient's capital base is thin (Source: forbes.com).

Models

Hierarchical Reasoning Model (HRM)

HRM is a 27-million-parameter model that reported performance on hard reasoning tasks (ARC-AGI-style puzzles, Sudoku, maze-solving) trained from only about 1,000 examples, without large-scale pretraining or chain-of-thought prompting. It uses a brain-inspired hierarchical architecture with two interdependent recurrent modules operating at different timescales, pitched as an alternative to chain-of-thought reasoning. HRM was open-sourced in 2025 and described in an arXiv paper (arXiv:2506.21734) (Source: arxiv.org).

Independent commentary questioned how much of the benchmark performance generalized beyond the narrow puzzle domains, and whether the architecture's gains came from the recurrent design or from task-specific training. As of this writing the reasoning claims remain unreplicated independently.

HRM-Text

On May 21–22, 2026, Sapient introduced HRM-Text, a fully open, 1.15-billion-parameter reasoning language model designed to reach benchmark performance using substantially less training data and compute than conventional language models (Source: sapient.inc). HRM-Text extends the HRM architecture from puzzle-solving to natural-language reasoning and is Sapient's first model positioned as a general-purpose, openly released language model rather than a research demonstrator.

ModelHRM-Text
ReleasedMay 2026
Parameters~1.15 billion
WeightsFully open
Design goalReasoning at low training-data / compute cost

Relationships

Sources

  • (Source: sapient.inc) — Sapient Intelligence, "Introducing HRM-Text" (May 2026).
  • (Source: forbes.com) — Forbes profile: Sapient Intelligence co-founded by Guan Wang in 2024, Singapore-based, AGI mission.
  • (Source: fortune.com) — Fortune (Nov 2025): founders William Chen and Guan Wang, OpenChat background, declined Musk funding.
  • (Source: arxiv.org) — "Hierarchical Reasoning Model," arXiv preprint (2025).

</parameter> </invoke>