Technology
Inherent says its AI agent outperforms frontier models in scientific replication
The London-based startup founded by DeepMind alumni claims its smaller agent outperformed systems from OpenAI and Anthropic at reproducing research.
The short version
- London startup Inherent announced that its AI agent, Faraday, outperformed larger models from OpenAI and Anthropic at independently replicating published research papers.
- Faraday operates on a 27-billion-parameter Qwen 3.6 model, utilizing reinforcement learning and OpenAI's GPT-5.5 Codex for coding tasks.
- The 12-person company recently raised a $50 million seed round and plans to expand its headcount to 25 by the end of the year as it pursues autonomous scientific discovery tools.
Key facts
- Inherent claims its AI agent, Faraday, outperformed OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8 at independently reproducing findings from scientific papers without prior knowledge of the results.[TechCrunch]
- Faraday is built on a 27-billion-parameter base model, Qwen 3.6, and uses reinforcement learning along with OpenAI's GPT-5.5 Codex for code execution.[TechCrunch]
- Inherent was founded by Google DeepMind alumni, recently emerged from stealth with a $50 million seed round, and currently employs about a dozen staff members in London.[TechCrunch]
- The startup plans to expand its team to roughly 20 to 25 employees by the end of the year.[TechCrunch]
What remains uncertain
- The benchmark data, full methodology, and independent verification of Faraday's performance claims against Anthropic and OpenAI models were not detailed in the report.[TechCrunch]
- It remains to be seen whether Faraday's replication capabilities can generalize into generating novel scientific discoveries as intended.[TechCrunch]