London-based AI lab Inherent, founded by former Google DeepMind researchers, has announced a significant breakthrough with its new AI agent, Faraday. Fresh off a $50 million seed round, the company revealed that its compact agent outperformed much larger models from industry leaders OpenAI and Anthropic. This achievement challenges the prevailing notion that superior AI performance requires massive computational scale.
A New Benchmark in AI Efficiency
Inherent's agent, Faraday, demonstrated its capabilities by independently reproducing the findings of published scientific papers, a complex task that serves as a training exercise for human scientists. In this specific challenge, it surpassed the performance of well-known frontier models like Anthropic's Claude Opus and OpenAI's GPT-5.5. The success signals a potential shift in AI development, where specialized systems can excel over generalist giants.
The most striking aspect of this achievement is Faraday's underlying architecture, which runs on a model with just 27 billion parameters. This is a small fraction of the size of the large-scale systems it competed against, which often contain hundreds of billions of parameters. This efficiency proves that targeted training can yield superior results without incurring the massive costs associated with larger models.
The Power of Reinforcement Learning
The key to Faraday's success lies in Inherent's training methodology, which heavily relies on reinforcement learning. Instead of being fed explicit rules, the AI is rewarded for positive outcomes, allowing it to develop what the company calls "research taste". This approach teaches the agent to intuitively identify which experiments are worth pursuing, mimicking the instincts of a seasoned researcher.
This training philosophy supports Inherent's ultimate ambition: to create an AI capable of making novel scientific discoveries. By fostering an ability to learn and adapt through rewards, the company aims to build an agent that can contribute meaningfully across multiple scientific fields. The focus is on creating a collaborative partner for human scientists rather than just a tool for verifying old data.
Strategic Focus and Future Ambitions
Inherent maintains a sharp focus on its core mission by strategically leveraging existing technologies for non-essential functions. For instance, Faraday utilizes OpenAI's coding tools, much like a human scientist would use established software instead of building everything from scratch. This practical approach allows the team to dedicate its resources to advancing the agent's unique scientific reasoning capabilities.
Situated in London's King's Cross AI hub, Inherent is poised for significant growth, planning to more than double its team size by the end of the year. The company's founders are committed to building a world-class research environment in the city, contributing to its reputation as a global AI center. With its recent funding and groundbreaking results, Inherent is becoming a compelling destination for top talent in the field.
Inherent's breakthrough with Faraday marks a potential turning point in the AI industry, demonstrating that smaller, highly specialized models can deliver superior performance. The company's innovative use of reinforcement learning to cultivate "research taste" paves the way for AI that acts as a true scientific collaborator. This achievement not only advances the quest for AI-driven discovery but also suggests a more efficient and accessible future for advanced AI applications.
Source: TechCrunch