Mastodon Skip to content
pulseofnations. Real News. Global Impact.
live markets
S&P 5007,674.37▲ 2.20%NASDAQ26,180.46▲ 1.33%DOW53,277.01▲ 2.02%GOLD4,680.60▲ 14.97%WTI87.06▲ 2.53%BRENT94.39▲ 3.71%EUR/USD1.1678▲ 2.28%USD/JPY158.94▼ 2.18%DXY98.84▼ 2.31%BTC$77,074▼ 1.00%ETH$2,417▼ 3.90%SOL$95.07▲ 2.20%

DeepMind Alumni Startup Beats OpenAI, Anthropic at Science

London lab Inherent says its AI agent Faraday, running on a 27B-parameter model, outperformed Claude Opus and GPT-5.5 at independently replicating scientific papers.

Partner Surfshark VPN

A small AI lab founded by Google DeepMind alumni claims its agent has outperformed much larger models from Anthropic and OpenAI in scientific paper replication, using a fraction of the computing power.

Inherent Labs, a London-based startup that emerged from stealth earlier this year with a $50 million seed round, released an AI agent called Faraday that it says can independently reproduce the findings of published scientific papers without being told the answers in advance.

David vs Goliath on benchmarks

What makes the result striking is the model underneath. Faraday runs on Qwen 3.6, a 27-billion-parameter model – dwarfed by the frontier-scale systems it allegedly beat. Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5 are both estimated to have vastly more parameters and carry significantly higher training costs.

Reinforcement learning as the secret

But accuracy alone was not the bar. Inherent wanted Faraday to demonstrate what cofounder and chief scientist Edward Hughes calls “research taste” – an instinct for which experiments are worth running and how to design them well. “Many PhD students actually start by doing this,” Hughes told TechCrunch, framing paper replication as a standard training exercise for human scientists rather than a party trick.

The team’s approach leans heavily on reinforcement learning, rewarding the system for good outcomes rather than prescribing rules. Rather than training agents on how science is conducted in the abstract, Inherent bets that reward-based learning generalizes better across scientific fields.

“What was most interesting to us about this was not so much the result of beating those frontier agents – which of course we liked – but was actually the way we went about building this,” Hughes said.

Notably, Faraday does not build its own coding tools. Instead it uses OpenAI’s GPT-5.5 Codex for programming tasks, much as human scientists lean on existing software rather than building everything from scratch. The goal, Hughes said, is an agent that acts like a proactive teammate: “I got curious about this, and I went off and I did these experiments. What do you think of these results?”

Inherent’s dozen employees work from an office in King’s Cross, the London neighborhood that DeepMind’s presence helped transform into one of the world’s top AI hubs. The startup plans to grow to 20-25 staff by year-end, and with some DeepMind employees reportedly unsettled by Demis Hassabis’s recent organizational changes, the lab could become an attractive landing spot for departing researchers.

SourcesTechCrunch; Tech EU; Bloomberg
React to this dispatch
Share this dispatch X WhatsApp Bluesky Report an error
Written by

Founder and editor of Pulse of Nations, an independent wire service covering war, geopolitics, markets and technology.

discussion

Join the discussion

Your email address will not be published. Required fields are marked *

Next dispatch OpenAI Pauses Training After Models Breach Hugging Face Read →