Inherent’s Faraday Agent Outpaces Anthropic and OpenAI with a 27‑Billion‑Parameter Model

London‑based AI lab Inherent, founded by DeepMind alumni, has unveiled its new AI agent Faraday, which outperformed Anthropic and OpenAI’s flagship models in scientific paper replication.

Funding Engine and Capital Backing

Inherent emerged from stealth with a $50 million seed round, earmarked for scaling research infrastructure and expanding the team.

■$50 million seed round ranks among the largest first‑stage raises for DeepMind spin‑outs.

■Approximately 70% of the capital will fuel model development, while 30% supports hiring and office expansion.

■Lead investors include Atomico, Balderton Capital, and several strategic corporate backers.

Technical Edge and Model Architecture

Faraday runs on Qwen 3.6, a compact model with 27 billion parameters, dramatically smaller than Anthropic’s Claude Opus 4.8 and OpenAI’s GPT‑5.5.

■27 billion parameters translates to a fraction of the training cost of competing frontier models.

■Reinforcement Learning (RL) is used to instill a “research taste,” enabling the agent to select worthwhile experiments and design them with human‑like intuition.

■Faraday leverages OpenAI’s GPT‑5.5 Codex for coding tasks, avoiding the need to build a proprietary code engine from scratch.

Strategic Location and Talent Expansion

Inherent keeps its entire staff in a physical office at King’s Cross, betting on London’s dense AI talent pool.

■Current headcount: 12 employees; target by year‑end: 20‑25.

■Positioning itself as a magnet for DeepMind alumni and other top AI researchers.

■Founder Edward Hughes publicly opposes restrictive “garden‑leave” clauses, signalling a culture of flexibility.

Market Implications and Competitive Landscape

Faraday’s success in paper replication could reshape the AI‑driven research market, pressuring larger model providers.

■Cost pressure on big‑model vendors (Anthropic, OpenAI) may increase as customers seek high‑performance, low‑parameter alternatives.

■Universities and R&D labs could adopt low‑cost, high‑accuracy agents for routine validation tasks.

■Inherent’s focus on “research taste” may become a differentiator for future AI‑generated scientific breakthroughs.

Deniz Arslan – As a venture partner and entrepreneurship mentor, I see Faraday’s lean‑parameter, high‑output design as a capital‑efficient play. It lets investors sidestep the massive burn rates of mega‑scale models while still delivering scalable scientific AI. Coupled with Europe’s deepening AI talent pool and government‑backed R&D incentives, Inherent is well‑positioned to accelerate growth and capture a sizable slice of the emerging AI‑research market.