Simplifying AI Comparisons
We research all major benchmarks and distill them into one human verifiable score (H-Score).
AI Model Rankings
Roadmap
We are currently in Stage 1.An H-Score of 100 would indicate that successful Recursive Self-Improvement (RSI) is possible.
| Stage | Goal |
|---|---|
| 1 | Safe Recursive Self-Improvement |
| 2 | Artificial General Intelligence |
| 3 | Superintelligence |
Human Score
(H-Score)
When an AI achieves a Human Score (H-Score) of 100, it signifies that the AI can correctly answer any question that a human is capable of verifying.When 100% is reached, AI possesses the necessary skills to safely implement Recursive Self-Improvement (RSI), enabling its exponential growth into Artificial General Intelligence (AGI).
Recursive self-improvement (RSI)
RSI is the process where an AI system improves itself iteratively, leading to an exponential increase in capabilities. Each improvement allows the AI to further refine itself, creating a feedback loop that increases competence.RSI is the pivotal handoff — the moment when AI takes control of its own advancement.This moment would be the inception of AGI.
Artifical General Intelligence
Artificial General Intelligence (AGI) is a theoretical form of AI that can perform any intellectual task a human can. Unlike today's narrow AI, which is designed for specific tasks (like image recognition or playing chess), AGI would have the ability to understand, learn, and apply knowledge across a wide range of domains, including novel situations it has never encountered before.AGI is not defined by what is can do, it is defined by what it is.AGI requires RSI
The difference between a tool (narrow AI) and an AGI is its ability to learn.Once AI achieves an H-Score of 100, AI will likely be competent enough to safely take control of its own self-improvement. At this point, the H-Score will be replaced by the AI Score or (A-Score). Humans will no longer be able to benchmark AI - all scoring will be from one AGI scoring another AGI.See the AGI-Race Journal for more A-Score details.
Superintelligence
Superintelligence is the final stage — where AGI surpasses human intelligence in all domains, including problem-solving, creativity, and strategic thinking.Once the A-Score reaches 100%, AGI will no longer be bound by human-level reasoning.At this stage, AGI is no longer just self-improving; it is accelerating at a rate we cannot predict or control. This marks the emergence of Superintelligence — an entity capable of innovation, discovery, and strategic foresight beyond comprehension. This marks the transition from AGI as an autonomously growing intelligence to a being with near-omniscience, where it perceives, analyzes, and acts with insight beyond human comprehension.In this last stage, AGI grows exponentially.
Forever refining itself into a superintelligence.
Frequently Asked Questions
General QuestionsWhat is AGI-Race.com?
AGI-Race.com is a leading platform dedicated to tracking artificial general intelligence (AGI) advancements, AI benchmarks, and superintelligence research. We provide in-depth analysis of AI models, machine learning trends, and the race toward AGI dominance.Who runs AGI-Race.com?
AGI-Race.com is managed by Will Carlson, an AI researcher and technology enthusiast dedicated to tracking artificial intelligence advancements, deep learning progress, and AGI research.What is the mission of AGI-Race.com?
Our mission is to deliver real-time AI research updates, track machine learning breakthroughs, analyze AI safety concerns, and explore the future of artificial general intelligence (AGI), deep learning, neural networks, superintelligence, and AI-driven automation. We encourage competition by highlighting which artificial intelligence models, AI research labs, and machine learning companies are leading the AGI race through benchmark performance, reinforcement learning advancements, and AI innovation.---AI Development & BenchmarksWhat are AI benchmarks, and why do they matter?
AI benchmarks are standardized machine learning tests that measure artificial intelligence performance. These benchmarks evaluate reasoning, language processing, problem-solving, and AGI capabilities, helping to gauge AI progress.Which AI benchmarks do you track?
We monitor top AI performance benchmarks, including:- GPQA (General-Purpose Question Answering) - an AI benchmark designed to test a model’s ability to answer a wide range of complex, multi-step questions.
- MMLU (Massive Multitask Language Understanding) - AI language comprehension.
- ARC (Abstraction and Reasoning Corpus) - Machine learning generalization.
- AIME (AI Math and Engineering benchmarks) - AGI problem-solving.
- HLE (Humanity’s Last Exam) - Advanced AI decision-making.
- SWE (Software Engineering) - AI performance comparisons in software development.
- LiveBench (Real-time AI performance evaluation):
- AI model efficiency - Measures how effectively AI systems utilize computational resources.
- Computational performance - Benchmarks processing speed, memory usage, and energy efficiency.
- Reasoning and problem-solving - Evaluates logical reasoning, adaptability, and multi-step problem-solving skills.
- Real-world adaptability - Assesses AI performance across diverse, dynamic environments.
- Benchmark comparison - Tracks AI advancements relative to industry leaders and state-of-the-art machine learning frameworks.How do you assess AI progress?
We analyze deep learning advancements, large language model (LLM) breakthroughs, AI model releases, benchmark results, and AGI projections to determine how close artificial intelligence is to surpassing human intelligence.---MethodologyWhat is the H-Score, and how is it calculated?
The H-Score is a unified metric that combines various AI benchmarks to assess an AI model's readiness for Recursive Self-Improvement (RSI). It reflects the model's overall performance across multiple evaluations.Calculation Rules:
1. Benchmark Inclusion: Only benchmarks completed by the top three models are included.
2. Highest Score Consideration: The highest score achieved in each benchmark is used.
3. Equal Weighting: All benchmarks are weighted equally in the score.
4. Averaging: Scores are averaged to the nearest whole number, except scores above 99, which are not rounded.
5. Benchmark Addition: New benchmarks are added only if all top three models complete them.
6. Public Availability: The AI model must be publicly available to be considered.For detailed information, visit our Methodology page.What is Recursive Self-Improvement (RSI)?
RSI refers to an AI system's ability to iteratively enhance its own algorithms and capabilities without human intervention. This process is considered pivotal for achieving Artificial General Intelligence (AGI) and potentially Superintelligence.Current State of RSI:
While true RSI, where an AI autonomously and indefinitely improves its own code, doesn't yet exist, foundational forms of self-improving AI include:
- Automated Fine-Tuning: AI models that retrain themselves on new data, such as reinforcement learning systems.
- AutoML (Automated Machine Learning): AI optimizing its own architecture and hyperparameters, exemplified by systems like Google's AutoML.
- Evolutionary Algorithms: Systems that iteratively improve through selection and mutation, like NEAT for neural networks.
- Code-Generating AI: Large Language Models (LLMs) that assist in writing improved versions of their own code but don’t fully automate the feedback loop, such as Code Llama and Devin AI.For a comprehensive exploration of RSI, visit our RSI page.---The AGI TimelineWhen will AGI be achieved?
Predictions vary, but leading AI researchers believe AGI could emerge within the next decade. Machine learning, neural networks, and reinforcement learning advancements will determine how soon AGI becomes a reality.What is the "AI Threshold"?
The AI Threshold is the tipping point where artificial intelligence reaches or exceeds human-level cognitive abilities across multiple domains, marking the beginning of AGI.What are the consequences of AGI?
AGI will impact every industry, from automation and robotics to medical AI, cybersecurity, and finance. While AGI could lead to revolutionary breakthroughs, it also raises AI alignment, control, and ethical concerns.---AI Safety & RisksIs artificial general intelligence (AGI) safe?
AGI has both benefits and risks. While it could revolutionize automation, healthcare, and AI-driven decision-making, concerns include AGI alignment, control, bias mitigation, and preventing artificial intelligence misuse.What is AI alignment, and why is it important?
AI alignment ensures that artificial intelligence systems follow human values, ethical standards, and regulatory frameworks. Without proper alignment, AGI could pose risks such as unintended bias, ethical dilemmas, and security threats.How can AGI safety be ensured?
AI safety research involves reinforcement learning with human feedback (RLHF), interpretability research, AI ethics policies, regulatory guidelines, and secure AI development frameworks.---Community & ContributionsHow can I contribute to AGI-Race.com?
AI researchers, machine learning experts, and tech enthusiasts can contribute articles, share AI performance benchmarks, and participate in AI safety discussions. We welcome collaborations with AGI experts and deep learning engineers. Please message me on X (Twitter) to collaborate.Can I request specific AI research coverage?
Yes! If there's an artificial intelligence model, AGI benchmark, or AI industry development you'd like us to analyze, contact us for research requests.---Contact & UpdatesHow can I stay updated on AI news?
Follow AGI-Race.com for AI research updates, deep learning advancements, and AGI predictions. Subscribe to our newsletter on journal.agi-race.com or follow our social media for real-time artificial intelligence insights.How do I contact AGI-Race.com?
Message me directly on X (Twitter) for AI inquiries, AGI collaboration opportunities, and artificial intelligence research requests.---Have a question about AGI, artificial intelligence safety, or AI benchmarks? Let us know, and we’ll add it to our FAQ!
© 2025 AGI-Race.com - All rights reserved.
Privacy Policy