Inherent, a London AI lab founded by former Google DeepMind employees, has announced results for an AI agent called Faraday: it successfully reproduced findings from published scientific papers without knowing the answers in advance and, by the company’s own evaluation, outperformed far larger models from Anthropic and OpenAI.
The most notable factor is its size. While its rivals are two frontier-scale systems, Faraday runs on Qwen 3.6 with a mere 27 billion parameters. Roughly speaking, parameter count typically corresponds to a model’s size and training compute. This means a system dozens of times smaller prevailed in a task requiring genuine research reasoning.
Inherent itself set a benchmark for success higher than conventional accuracy. In addition to reproducing correct results, the company wanted Faraday to demonstrate what it calls “research taste”: the instinct to recognize which experiments are worth running and how best to design them. The method used to instill this intangible quality is reinforcement learning, an approach that rewards good outcomes rather than relying on predefined rules. Inherent did not train the agent primarily on knowledge of how science operates, betting instead that a reward-based approach would generalize better toward its long-term goal: an agent contributing across diverse scientific disciplines.
Co-founder and chief scientist Edward Hughes said paper replication is not a parlor trick. “Many graduate students actually start with this.” What resonates most with him is not simply defeating frontier agents, but the path taken to get there. He also emphasized a collaborative philosophy: rather than an agent telling users what they want to hear, the goal is the kind of colleague who returns and says, “I was curious about this, so I ran a few experiments—take a look at the results.”
This frugal approach is also visible in its infrastructure: instead of building custom coding tools from scratch, Faraday simply uses OpenAI’s GPT-5.5 Codex, much like human scientists rely on readily available software.
A few figures regarding the company: Inherent emerged from stealth in May with a $50 million seed round, currently employs around 12 people working on-site at its King’s Cross office, and plans to grow to 20–25 by the end of the year. Its four co-founders are Louis Kirsch, Kaloyan Aleksiev, Tantum Collins, and Edward Hughes.
Important note: The comparative results above were published by Inherent itself and have not been independently verified by a third party.