DeepMind Alumni Startup Inherent Claims AI Agent Beats Anthropic and OpenAI at Research Replication
Original: Inherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research
Why This Matters
Demonstrates that smaller, efficiently trained AI models can match or exceed frontier-scale systems in specialized scientific tasks.
London-based AI lab Inherent, founded by Google DeepMind alumni, says its AI agent Faraday outperformed Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at independently replicating published scientific research, using a 27-billion-parameter model—a fraction of the competing systems' scale.
Inherent, a London AI startup founded by Google DeepMind alumni and backed by a $50 million seed round, has released its AI agent Faraday, claiming it outperformed frontier models from Anthropic and OpenAI on a scientific paper replication benchmark. Faraday runs on Qwen 3.6, a relatively compact 27-billion-parameter model, while its competitors—Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5—operate at significantly larger scale. The benchmark task required agents to independently reproduce the findings of published scientific papers without being given the answers in advance. Cofounder and chief scientist Edward Hughes emphasized that beating rivals was secondary to the methodology: 'What was most interesting to us was not so much the result of beating those frontier agents, but was actually the way we went about building this.' Inherent uses reinforcement learning—rewarding agents for good outcomes rather than prescribing explicit rules—to train Faraday to develop 'research taste,' defined as an instinct for which experiments are worth running and how to design them. The company's long-term goal is to build AI capable of generating new scientific knowledge across multiple fields. Notably, Faraday relies on OpenAI's GPT-5.5 Codex for coding tasks rather than building its own tool, reflecting a focused approach to system design.