Inherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research

Inherent, a British AI laboratory founded by former DeepMind researchers, has announced that its newly developed AI agent, Faraday, has achieved superior performance compared to systems from industry leaders Anthropic and OpenAI in replicating scientific research. This breakthrough suggests that specialized AI agents designed for scientific tasks may represent a significant advancement in automating complex research workflows and accelerating innovation across multiple disciplines.

Inherent's Faraday agent demonstrated enhanced capabilities in systematically replicating published scientific papers, outperforming Claude (Anthropic's flagship model) and GPT-4 (OpenAI's leading offering) in benchmarking tests. The system was specifically engineered to handle the multifaceted challenges inherent in scientific reproduction, including experimental design interpretation, methodology implementation, and results validation. By focusing on this specialized domain, Inherent has positioned its technology as a purposeful alternative to general-purpose large language models, addressing gaps in existing AI systems when applied to scientific contexts.

• Scientific research acceleration could be significantly enhanced through automated paper replication, reducing time-to-insights for researchers
• Specialized AI agents may prove more effective than general-purpose models for complex domain-specific tasks
• The breakthrough validates the business model of focused AI development rather than competing solely on general capabilities
• Potential applications span pharmaceuticals, materials science, and climate research where reproducibility remains a critical bottleneck
• Competition in AI is shifting from raw model scale to specialized performance capabilities
• Research institutions may reduce dependency on major AI companies by adopting purpose-built solutions

Inherent's success with Faraday demonstrates an important market opportunity: specialized AI agents addressing specific professional challenges may outcompete general-purpose systems in their target domains. For the scientific community, this development promises to accelerate research validation, reduce publication bias from non-reproducible results, and democratize access to advanced research capabilities. As AI becomes increasingly integrated into professional workflows, the ability to develop focused solutions tailored to industry-specific needs will likely determine competitive advantage. This news signals that the future of AI may belong as much to specialized innovators as to generalist tech giants.