A 27-billion-parameter model beat Opus 4.8 and GPT-5.5 at reproducing published papers
Inherent, a London lab of about a dozen people founded by Google DeepMind alumni, said on 22 August that its agent Faraday produces more faithful replications of published scientific findings than Anthropic's Claude Opus 4.8 or OpenAI's GPT-5.5. Faraday runs on Qwen 3.6, an open 27-billion-parameter model, and was trained with long-horizon reinforcement learning. The test is Replica, Inherent's own suite of 310 tasks drawn from 100 papers in natural-language processing, materials science and weather forecasting; each task asks the agent to reproduce a figure without ever seeing the original plot, under limited time and compute. Inherent reports leading in every category, with the largest margins in meta-learning, structural biology and materials science. The lab left stealth in May 2026 with a $50 million seed round and has not yet shown a genuinely new discovery.
Related stories
Alibaba trained a model on a rack of 100 real phones — it now finishes 92 out of 100 tasks on an actual handset2026-08-23OpenAI has given away the engine under Codex, and its own numbers say the engine is worth more than the model
2026-08-22Give a model the reference list of an unpublished paper and ask what the paper says. The best ones get it 15 percent of the time.
2026-08-22Google's open models have been downloaded a billion times, and outsiders have built 100,000 versions of them
2026-08-22Three-quarters of Americans do not want a data centre near them. A year ago they were evenly split.
2026-08-22