New AI Player 'Inherent' Stuns Rivals, Outperforms Anthropic and OpenAI

Inherent, a dynamic London-based AI laboratory founded by esteemed Google DeepMind alumni, has recently captured attention in the artificial intelligence community. After emerging from stealth mode with a substantial $50 million seed round, the company has begun to unveil its groundbreaking work, notably its AI agent named Faraday. This development comes at a time when other well-funded rivals from similar backgrounds have yet to present concrete innovations to the public.

Faraday has achieved a significant milestone by outperforming much larger and more established models, such as Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, in a highly specific and complex task: independently reproducing the findings of published scientific papers. What makes this achievement particularly remarkable is that Faraday accomplished this feat using a comparatively tiny model called Qwen 3.6, which possesses only 27 billion parameters. This is a stark contrast to the frontier-scale systems it surpassed, highlighting an efficiency that could prove highly appealing to investors, as model size and training costs are often correlated with the number of parameters.

While replicating scientific papers might seem like a niche application, co-founder and chief scientist Edward Hughes emphasizes its foundational importance, likening it to a standard training exercise for human PhD students. However, Inherent's ultimate goal extends far beyond mere verification of old results; their ambitious "north star" is to build AI capable of discovering new scientific knowledge. The significance, according to Hughes, lies not just in beating frontier agents, but in the innovative methodology employed to achieve it.

Inherent’s approach prioritizes imbuing its AI agents with "research taste"—an intangible yet critical instinct for discerning which experiments are truly worth pursuing and how to design them effectively. This challenging aspect of AI training is tackled through reinforcement learning, a reward-based method that encourages desirable outcomes rather than providing explicit rules. By focusing on this reward-centric approach, Inherent believes its agents will generalize better, enabling contributions across a wide array of scientific fields and aligning with its long-term vision of developing a true AI scientist.

The company's strategic choices also reflect a pragmatic and collaborative philosophy. Rather than developing proprietary coding tools, Faraday leverages existing software like OpenAI’s GPT-5.5 Codex, mirroring how human scientists utilize established resources. Furthermore, Inherent aims to foster a collaborative dynamic with its AI, designing agents that are curious and experimental, much like an ideal human teammate who independently explores ideas and presents findings for discussion, rather than simply confirming preconceived notions.

Culturally, Inherent champions an in-person work environment, with its dozen employees collaborating from an office in King's Cross, London. This area, transformed into a leading AI hub partly due to Google DeepMind’s presence, is seen by Hughes as the ideal location due to its density of AI talent. Edward Hughes has also voiced a personal stance against "garden leave," a common U.K. practice that restricts departing employees from joining rival companies for months, a constraint he personally encountered. He argues this practice hinders British startups compared to their American counterparts who face fewer such restrictions.

Looking ahead, Inherent is poised for significant expansion, with plans to grow its headcount to between 20 and 25 employees by the end of the year. Given its ambitious objectives in world models and the potential for staff shifts at DeepMind following Demis Hassabis’s new role, Inherent's growth trajectory could position it as an attractive destination for top AI talent. The company was founded by Edward Hughes, alongside Louis Kirsch, Kaloyan Aleksiev, and Tantum Collins, all committed to advancing the frontier of AI for scientific discovery.