AI Brain Model Outperforms ChatGPT in Reasoning

Here’s a breakdown of the‍ key information from the provided text:

HRM Model: A new AI model with 27 million parameters trained on only 1,000 samples.This is significantly smaller‍ than⁣ most advanced Large Language Models (llms).
LLM ⁣Comparison:
GPT-5 (estimated): 3-5 trillion parameters.
Other LLMs: Generally have billions or trillions of‍ parameters.
ARC-AGI ‍Benchmark Results: HRM performed surprisingly well on the⁢ challenging ARC-AGI benchmark, outperforming other models:
ARC-AGI-1: HRM – 40.3%, OpenAI’s o3-mini-high – 34.5%, Anthropic’s Claude 3.7 – ‍21.2%, ⁢Deepseek R1 – 15.8%

ARC-AGI-2: HRM – 5%, o3-mini-high – 3%, Deepseek⁤ R1 ⁤- ⁣1.3%, Claude 3.7 – 0.9%
Reasoning Approach: Most LLMs⁣ use “chain-of-thought” (CoT) reasoning, breaking down problems into smaller steps. The article doesn’t explicitly state how HRM ⁤reasons, but its⁤ success despite its small size suggests a possibly different approach.

In essence, the article highlights a ⁢new AI model (HRM) that achieves extraordinary results despite being much smaller⁤ than current leading LLMs, suggesting a potentially innovative approach to AI progress.