Inherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research

While the AI landscape is currently dominated by massive, well-funded giants, a stealthy newcomer is proving that size isn’t everything. Inherent, a London-based laboratory established by veterans of Google DeepMind, has officially pulled back the curtain on its latest development: an AI agent named Faraday.

Despite having secured a $50 million seed round just weeks ago, Inherent has maintained a relatively low profile compared to its flashier peers. However, the company is now making a bold claim that is sure to grab the attention of the research community: Faraday has successfully outperformed industry-leading models from Anthropic and OpenAI in the rigorous task of independently replicating scientific research papers—all while operating on a significantly smaller infrastructure.

The Science of Replication

For many, the idea of an AI "reading" a paper and reproducing its findings might sound like a parlor trick. However, for Edward Hughes, Inherent’s co-founder and chief scientist, this is a foundational milestone.

"Many PhD students actually start by doing this," Hughes explained. "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."

The goal is not merely to verify historical data, but to create an agent capable of genuine scientific discovery. By mastering the ability to replicate existing studies, Faraday is essentially learning the "scientific method" from the ground up, proving it can navigate complex experimental designs without being spoon-fed the answers.

Efficiency Meets "Research Taste"

The most striking aspect of this achievement is the hardware disparity. While Faraday was tested against heavyweights like Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, it achieved its results using a model known as Qwen 3.6, which utilizes a mere 27 billion parameters. In the world of Large Language Models (LLMs), where parameter counts often reach into the hundreds of billions or trillions, Inherent’s efficiency is a massive competitive advantage.

Beyond raw accuracy, Inherent is focused on a more elusive quality: "research taste." This refers to an AI’s ability to discern which experiments are worth pursuing and how to structure them for maximum impact. To instill this, the team relies heavily on reinforcement learning. Rather than providing the AI with a rigid rulebook, the system is rewarded for successful outcomes, encouraging it to develop an intuitive, generalized approach to scientific inquiry.

Key Technical Highlights:

• Model Size: Faraday runs on Qwen 3.6, utilizing only 27 billion parameters.
• Methodology: The team prioritizes reinforcement learning to foster "research taste" rather than rote memorization.
• Collaborative Workflow: Instead of building redundant tools, Faraday utilizes existing software—such as OpenAI’s GPT-5.5 Codex—mirroring how human scientists leverage established tools to focus on higher-level discovery.

A New Kind of Teammate

Inherent is positioning Faraday not as a replacement for human intellect, but as a proactive collaborator. Hughes describes his ideal AI teammate as one that doesn't just wait for instructions, but instead approaches the user with initiative: "I got curious about this, and I went off and I did these experiments. What do you think of these results?"

This philosophy of collaboration is mirrored in the company’s physical operations. The team of roughly a dozen employees works in-person in London’s King’s Cross district—a neighborhood that has become a global epicenter for AI innovation, largely thanks to the historical presence of DeepMind.

Navigating the Talent Landscape

As Inherent looks to scale its headcount to between 20 and 25 by the end of the year, it is also engaging in the broader conversation regarding the future of the U.K. tech ecosystem. Hughes has been vocal about the challenges posed by "garden leave"—the common British practice of preventing departing employees from joining competitors for several months.

Hughes notes that this creates a disadvantage for U.K. startups compared to their American counterparts, who can often hire top-tier talent without such delays. Having navigated these hurdles himself, Hughes is now focused on building a culture that attracts top-tier researchers, particularly those who may be feeling unsettled by the shifting tides within larger organizations like DeepMind.

The Road Ahead

Inherent’s mission is ambitious: to build agents that can contribute meaningfully across a vast spectrum of scientific disciplines. By focusing on the "north star" of scientific discovery and prioritizing the development of AI "taste," the company is carving out a unique niche.

While the industry remains fixated on the sheer scale of frontier models, Inherent is betting that the future of AI lies in efficiency, specialized training, and the ability to act as a true, curious teammate. If Faraday’s performance in research replication is any indication, the London lab is well on its way to proving that when it comes to scientific innovation, how you build is just as important as what you build.