Inherent, a London-based AI lab founded by Google DeepMind alumni, has unveiled its new AI agent, Faraday. In a recent test, Faraday outperformed much larger models from Anthropic and OpenAI in replicating research findings, using a fraction of the computational power.

Faraday runs on Qwen 3.6, a smaller model with just 27 billion parameters, compared to the vast GPT-5.5 and Claude Opus 4.8 systems from its competitors. Inherent’s success lies not only in accuracy but also in demonstrating 'research taste'—the instinct for choosing which experiments are worth running and designing them well.

Edward Hughes, cofounder and chief scientist of Inherent, emphasizes the importance of reinforcement learning, a training method that rewards good outcomes rather than prescriptive rules. This approach aims to enable agents capable of contributing across many scientific fields, moving beyond mere verification of existing knowledge.

In contrast to building its own tools from scratch, Inherent leverages existing software like OpenAI’s GPT-5.5 Codex, mirroring how human scientists utilize pre-existing technologies. Hughes also advocates for a more collaborative approach, likening his ideal teammate to one who returns with insights and experimental results after venturing off on their own.

The company’s focus on building an AI scientist agent imbued with 'taste' is reflected in its hiring strategy; Inherent is currently expanding its team of 12 employees based in King’s Cross, London, aiming for about 20 to 25 by the end of the year. This could make it a tempting destination for DeepMind employees considering a move.