BestLLMfor Your hardware. Your LLM. Your call.
The Local Copilot Kit APIOpen data Find my LLM
Model fiche

HRM-Text 1B

By Sapient · United States

Updated 2026-08-31

reasoning small
Parameters
1.2B
License
Apache 2.0
Context
2k
VRAM (Q4)
0.7 GB
Released
2026-05-17

Overview

⚠ Research model: Sapient's 1.2B Apache-2.0 prefix-LM built around Hierarchical Reasoning Model (HRM) architecture. Not instruction-tuned or chat-ready — intended for reasoning R&D, not production assistants.

When to pick this model

  • Studying hierarchical reasoning architectures as a research baseline
  • Prototyping HRM-based fine-tunes or reasoning experiments
  • Academic benchmarking against arXiv:2605.20613's reported results
  • Testing hierarchical planning approaches at small scale before scaling up

VRAM requirements by quantization

VRAM REQUIRED (GB)Q4_K_M0.7 GBQ5_K_M0.9 GBQ8_01.3 GBFP162.4 GB
QuantizationVRAM required
Q4_K_M (recommended)0.7 GB
Q5_K_M0.9 GB
Q8_01.3 GB
FP16 (no quantization)2.4 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

In practice, HRM-Text 1B fits an 8 GB consumer card at Q4_K_M (0.7 GB). Stepping up to Q8_0 nearly doubles the footprint to 1.3 GB, and unquantized FP16 weights take 2.4 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, HRM-Text 1B needs roughly 1.6 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 45 tokens/sec on entry-level GPUs, on the order of 110 tokens/sec on a mid-range card, and up to 220 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches HRM-Text 1B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memoryExample cardsBest fit for HRM-Text 1B
8 GBRTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GBFP16 (2.4 GB used)
12 GBRTX 5070, RTX 5070 Ti Laptop, RTX 4080 LaptopFP16 (2.4 GB used)
16 GBRTX 5080, RTX 4080 Super, Radeon RX 9070 XTFP16 (2.4 GB used)
24 GBRTX 4090, Radeon RX 7900 XTX, RTX 5090 LaptopFP16 (2.4 GB used)
32 GBRTX 5090FP16 (2.4 GB used)

Which hardware should you buy to run HRM-Text 1B?

To run HRM-Text 1B locally at Q4, you need ~0.7 GB of VRAM. The best value for this today is a RTX 5060 Ti 16GB (ASUS Dual OC) (16 GB VRAM, best $/GB).

Check RTX 5060 Ti 16GB (ASUS Dual OC) price on Amazon →

As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.

Strengths

  • Apache 2.0 license allows unrestricted commercial use
  • Experimental HRM architecture designed for reasoning tasks
  • Extremely compact at 1.2B parameters
  • Useful reference implementation for reasoning-architecture research

Limitations

  • Not instruction-tuned or chat-aligned — unusable as a drop-in assistant
  • English only
  • Thin documentation and an early-stage ecosystem

Typical workloads

In our catalog grid, HRM-Text 1B is filed under Research, Hierarchical Reasoning, Pre-Alignment Study — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: multi-step reasoning and math-flavoured tasks.

Note the 2k-token context window — fine for short interactions, limiting for long documents or big retrieval contexts. The Apache 2.0 license is permissive, so shipping it inside a commercial product raises no special legal questions.

Architecture & training

Architecture: Prefix-LM transformer · 1.2B parameters · Hierarchical Reasoning Model (HRM)

Training: English pretraining. A pre-aligned research model, not instruction-tuned, focused on hierarchical reasoning (paper arXiv:2605.20613).

Verdict

A research curiosity for hierarchical reasoning architectures, not a model you'd deploy as a chat assistant.

Quick start

# HuggingFace : sapientinc/HRM-Text-1B

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.

Similar models worth comparing

Frequently asked questions

How much VRAM does HRM-Text 1B need?

At the recommended Q4_K_M quantization, HRM-Text 1B needs about 0.7 GB of VRAM. Q8_0 takes 1.3 GB, and unquantized FP16 weights take 2.4 GB.

Can HRM-Text 1B run without a GPU?

Yes — with roughly 1.6 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does HRM-Text 1B support?

HRM-Text 1B supports a 2k-token context window (2,048 tokens).

Can I use HRM-Text 1B commercially?

Yes. HRM-Text 1B is released under Apache 2.0, a permissive open-source license that allows commercial use, modification and redistribution.

How fast is HRM-Text 1B on consumer hardware?

Our compatibility engine estimates on the order of 110 tokens/sec on a mid-range GPU and up to 220 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of HRM-Text 1B should I download first?

Start with Q4_K_M (0.7 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It already fits an 8 GB card at FP16.

Tools

Is HRM-Text 1B the right pick for you?

Compute self-hosted ROI → Back to catalog