Training a large language model from scratch is supposed to be expensive. Sapient Intelligence just did it for less than the cost of a MacBook Pro.
The Singapore-based startup released HRM-Text, a 1.15-billion-parameter language model trained on 16 GPUs over 1.9 days at a total cost between $1,000 and $1,500. The model is fully open-sourced on GitHub and Hugging Face, which means anyone can inspect, modify, and deploy it.
How HRM-Text works, and why it matters
Traditional Transformer-based models, the architecture behind GPT and its cousins, typically require training on trillions of tokens. HRM-Text was trained on roughly 40 billion structured tokens. That’s orders of magnitude less data, yet the model still posts competitive benchmark scores.
On the MATH benchmark, HRM-Text scored 56.2. On DROP, a reading comprehension test that requires discrete reasoning, it hit 82.2. Sapient positions these results against models like Meta’s Llama 3.2 3B and Alibaba’s Qwen 3.5 2B, both of which required substantially more resources to train.
The company behind the model
Sapient Intelligence was founded in 2024 by Guan Wang and William Chen. The company raised a $22 million seed round in January 2025, pushing its valuation past $200 million.
The HRM architecture itself debuted in a June 2025 paper, where Sapient demonstrated competitive performance using a model with just 27 million parameters. HRM-Text scales that approach by roughly 40x in parameter count while keeping compute costs negligible by industry standards.
What this means for crypto and decentralized AI
One of the biggest bottlenecks for on-chain AI inference is compute cost. Running a multi-billion-parameter model on decentralized GPU networks like Akash, Render, or io.net is expensive and slow. A model that achieves meaningful reasoning at 1.15 billion parameters, trained on a fraction of the typical data, suddenly becomes a much more realistic candidate for decentralized deployment.
A fully open-source model architecture that anyone can train for $1,500 aligns naturally with the ethos of decentralized networks that want to offer AI services without depending on OpenAI or Anthropic APIs.