DeepMind Alumni’s AI Agent Just Outperformed Anthropic and OpenAI
A London AI lab founded by Google DeepMind alumni just made some noise. Their AI agent, Faraday, outperformed much larger models from Anthropic and OpenAI at a specific task.
And they did it with a fraction of the size.
Let me break down what they built.
The Company
Inherent is a London AI lab founded by Google DeepMind alumni. They just emerged from stealth with a $50 million seed round.
The team is small. About a dozen employees all work in person out of an office in King’s Cross, the London neighborhood that DeepMind’s presence helped turn into one of the world’s top AI hubs.
They’re planning to grow to “about 20 to 25” by the end of the year.
The Agent: Faraday
Inherent’s AI agent is called Faraday. And it just outperformed larger, better-known models at a specific task: independently reproducing the findings of published scientific papers without being told the answer in advance.
Measured against Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, Faraday runs on a comparatively tiny model called Qwen 3.6 that has just 27 billion parameters.
For context, those are frontier-scale systems with much larger parameter counts.
A mysterious AI model called Ox Alpha recently appeared on OpenRouter, sparking intense speculation about who built it. Read more about the Ox Alpha mystery here.
Why This Matters
Beating other AI systems at paper replication wasn’t the point. How they got there was.
Cofounder and chief scientist Edward Hughes put it this way:
“What was most interesting to us about this was not so much the result of beating those frontier agents — which of course we liked — but was actually the way we went about building this.”
Beyond replicating results, Inherent wanted Faraday to demonstrate “research taste.” That’s an instinct for what experiments are worth running and how to design them well.
The Secret: Reinforcement Learning
Teaching something as intangible as taste is hard. That’s where reinforcement learning comes in.
It’s a training method that rewards an AI system for good outcomes rather than spelling out rules for it to follow. Inherent leans on this reward-based approach, betting it will generalize better to their longer-term goal of agents capable of contributing across many scientific fields.
“We’re always guided by that north star of building an AI scientist agent and imbuing our agents with taste.”
The Collaborative Approach
Inherent is trying to avoid building agents that simply tell users what they want to hear.
Instead, Hughes said the goal is modeled on his favorite kind of teammate:
“I got curious about this, and I went off and I did these experiments. What do you think of these results?”
Rather than developing its own coding tool, Faraday uses OpenAI’s GPT-5.5 Codex instead. Much the way human scientists lean on existing software rather than building everything themselves.
The London Factor
Hughes is bullish on London’s density of AI talent. But he’s also calling to end “garden leave.” That’s the practice in the UK of barring departing employees from joining or starting a rival company for months after they resign.
American researchers generally don’t face this restriction, giving US startups a head start on hiring talent.
“This is a personal view rather than a company view, but I was affected by the garden leave problem.”
Hughes eventually got around that constraint and started Inherent alongside three other cofounders.
The Bottom Line
Inherent’s AI agent Faraday outperformed Anthropic and OpenAI models at replicating scientific papers. It runs on a much smaller model with 27 billion parameters. The company is using reinforcement learning to build an AI scientist with “research taste.” And they’re hiring.
This is one to watch.
