Faraday’s Small Model, Big Impact: AI Agent Outshines Anthropic and OpenAI

Inherent, the London artificial intelligence lab founded by former Google DeepMind researchers, announced that its newly released agent Faraday outperformed the larger models of Anthropic and OpenAI.

Small Model, Big Impact: Faraday’s 27‑Billion‑Parameter Edge

■$50 million seed round

■Faraday runs on the Qwen 3.6 model with 27 billion parameters

■Beats Anthropic’s Claude Opus 4.8 and OpenAI’s GPT‑5.5, both frontier‑scale systems

■Aims to grow headcount to 20‑25 by year‑end

Research Taste and Reinforcement Learning: How It Was Built

■Trained with reinforcement learning that rewards successful outcomes instead of prescribing explicit rules

■Uses OpenAI’s GPT‑5.5 Codex for coding tasks, mirroring how scientists rely on existing software

■Seeks to demonstrate “research taste” – the instinct to choose worthwhile experiments and design them well

■This approach supports the long‑term goal of an agent that can contribute across many scientific fields

London’s AI Hub: King’s Cross Office and Future Plans

■12 employees work in person from the King’s Cross office

■Cofounders: Louis Kirsch, Kaloyan Aleksiev, Tantum Collins, and Edward Hughes

■Hughes has spoken out against the UK’s “garden leave” practice, advocating its abolition

■The office sits in King’s Cross, a district transformed into a leading AI hub by DeepMind’s presence

Gökberk Uçar: An agent like Faraday that exhibits scientific‑discovery ability could accelerate data‑intensive analyses in air‑freight pricing and operational‑margin models. Its small‑model, high‑performance profile offers cost advantages for compute‑heavy logistics tasks such as capacity and route optimization, potentially boosting carriers’ competitive edge.