Editorial illustration for Tiny AI Model TRM Outperforms GPT-4o on Complex Reasoning Test
Tiny AI Model Beats GPT-4o in Complex Reasoning Challenge
Tiny AI Model TRM Beats GPT-4o and Gemini 2.5 Pro on ARC-AGI Test
Everyone in AI is building a bigger model. A team in Montreal just built a smarter one, and it’s microscopic.
The Tiny Recursive Model from Samsung’s SAIL lab has seven million parameters. That’s a rounding error. GPT-4o has hundreds of billions.
Yet on the famously tricky ARC-AGI benchmark, which tests abstract reasoning, the tiny model won. It scored 45% on ARC-AGI-1 and 8% on ARC-AGI-2. OpenAI’s o3-mini-high managed 3%.
Gemini 2.5 Pro got 4.9%. Claude 3.7 scraped 0.7%.
TRM uses less than one ten-thousandth of the computational power of those giants. It doesn’t just answer questions. It solves Sudoku puzzles and visual logic tests by thinking in loops, running its simple network over and over to reason step-by-step.
A new mini-model called TRM shows that recursive reasoning with tiny networks can outperform large language models on tasks like Sudoku and the ARC-AGI test - using only a fraction of the compute power. Researchers at Samsung SAIL Montreal introduced the "Tiny Recursive Model" (TRM), a compact design that outperforms large models such as o3-mini and Gemini 2.5 Pro on complex reasoning tasks, despite having just seven million parameters. By comparison, the smallest language models typically range from 3 to 7 billion parameters.
According to the study "Less is More: Recursive Reasoning with Tiny Networks," TRM reaches 45 percent on ARC-AGI-1 and 8 percent on ARC-AGI-2, outperforming much larger models including o3-mini-high (3.0 percent on ARC-AGI-2), Gemini 2.5 Pro (4.9 percent), DeepSeek R1 (1.3 percent), and Claude 3.7 (0.7 percent). The authors say TRM achieves this with less than 0.01 percent of the parameters used in most large models. More specialized systems such as Grok-4-thinking (16.0 percent) and Grok-4-Heavy (29.4 percent) still lead the pack.
The result is a direct challenge to the entire industry’s religion. For a decade, the credo has been simple: more parameters equal more intelligence. This is a seven-million-parameter rebuttal.
It proves that raw scale can be a crutch. Clever design, it turns out, matters more.
Specialized systems like Grok-4 still post higher scores. But the race is no longer just about who can stack the most silicon. It’s about who can write the most intelligent loop.
The future of AI might not be a single, monstrous brain. It could be a swarm of tiny, persistent ones.
Related Reading
Browse all Google news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
Related Topic Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks
From Archives WIRED Roundup: Zoë Schiffer on the AI Bubble and This Week's Top Stories
From Archives Raindrop Tackles AI Agent Regressions With New Experimentation Platform
From Archives Google's 1.3 Quadrillion AI Tokens Show Compute Scale, Not Business Value
Business & Startups
From Archives Google Targets Enterprise AI Market with New Tools
Business & Startups
Common Questions Answered
How does the Tiny Recursive Model (TRM) challenge existing assumptions about AI model performance?
The TRM demonstrates that compact AI systems with only seven million parameters can outperform large language models like GPT-4o and Gemini 2.5 Pro on complex reasoning tasks. This breakthrough suggests that recursive reasoning and model efficiency might be more important than raw computational scale in AI development.
What specific complex reasoning tests did the TRM successfully complete?
The TRM showed exceptional performance on challenging tasks like Sudoku and the ARC-AGI test, which traditionally require sophisticated reasoning capabilities. By excelling in these tests, the tiny model from Samsung SAIL Montreal proved that small neural networks can solve intricate problems more efficiently than much larger models.
Why is the seven-million-parameter TRM considered significant in AI research?
The TRM represents a potential paradigm shift in AI model design by proving that smaller, more focused models can achieve superior performance through recursive reasoning techniques. Its success challenges the long-held belief that larger models with more parameters are inherently more capable of complex computational tasks.
Further Reading
- Papers with Code - Latest NLP Research - Papers with Code
- Hugging Face Daily Papers - Hugging Face
- ArXiv CS.CL (Computation and Language) - ArXiv
