London-based AI startup Inherent has revealed Faraday, an AI agent that outperforms massive models from OpenAI and Anthropic at reproducing scientific research.
- Inherent, founded by former Google DeepMind employees, has launched its AI agent 'Faraday'.
- Faraday outperformed frontier models from OpenAI and Anthropic in reproducing scientific paper findings.
- The agent operates on a significantly smaller 27-billion parameter model (Qwen 3.6).
In a significant breakthrough for the artificial intelligence sector, London-based startup Inherent has announced that its AI agent, Faraday, has successfully outperformed much larger, more resource-intensive models from industry giants Anthropic and OpenAI. The specialized task? Independently reproducing the findings of published scientific papers without prior knowledge of the answers.
Emerging from stealth shortly after securing a $50 million seed round, Inherent is positioning itself as a specialized player in the scientific AI space. While many startups focus on general-purpose chatbots, Inherent is targeting the creation of an 'AI scientist.' Co-founder and Chief Scientist Edward Hughes noted that replicating research is a fundamental training step for human scientists, making it the perfect benchmark for an intelligent agent.
Why This Matters
BozokMedia analysis shows that the efficiency of Faraday represents a paradigm shift in AI development. Instead of relying on sheer scale, Inherent is focusing on 'research taste'—the ability to discern which experiments are worth pursuing. This suggests that the next era of AI will be defined by specialized reasoning rather than just massive parameter counts.
The goal is not just to beat frontier agents, but to build an agent that possesses the instinctual 'taste' of a human researcher.
What makes this achievement truly remarkable is the scale of the underlying technology. While models like OpenAI's GPT-5.5 and Anthropic's Claude Opus require massive computational power, Faraday runs on Qwen 3.6, a relatively compact model with only 27 billion parameters. By utilizing reinforcement learning, Inherent has trained its agent to learn from outcomes rather than just following static rules, allowing for better generalization across scientific disciplines.
The company operates out of King’s Cross, London, a burgeoning hub for AI talent. Hughes has also been vocal about the challenges facing UK talent, specifically the 'garden leave' practice that limits the mobility of researchers. Despite these hurdles, Inherent is on an aggressive growth trajectory, planning to expand its team to 25 employees by the end of the year.
| Feature | Faraday (Inherent) | Frontier Models (OpenAI/Anthropic) |
|---|---|---|
| Model Size (Approx) | 27 Billion Parameters | Trillions of Parameters |
| Primary Focus | Scientific Research Replication | General Purpose Intelligence |
| Core Methodology | Reinforcement Learning / Research Taste | Massive Scale Pre-training |
Frequently Asked Questions
1. What is 'research taste' in the context of Faraday?
It refers to the AI's ability to understand which scientific experiments are most valuable and how to design them effectively.
2. Where is Inherent located?
Inherent is based in London, specifically in the King's Cross area, which is a major global AI hub.
