llms.blog
AI

Inherent Releases Faraday: 27B Scientific Agent Outperforms Frontier Models on Paper Replication

London-based AI research startup Inherent has released Faraday, an autonomous AI agent engineered to independently reproduce published scientific research without prior exposure to target solutions. Founded by former Google DeepMind researchers Louis Kirsch, Kaloyan Aleksiev, Tantum Collins, and Edward Hughes, the lab launched Faraday weeks after securing a $50 million seed round. According to benchmark results published by the lab, Faraday outperformed significantly larger frontier systems, in

axc
axc
2 min
Inherent Releases Faraday: 27B Scientific Agent Outperforms Frontier Models on Paper Replication

London-based AI research startup Inherent has released Faraday, an autonomous AI agent engineered to independently reproduce published scientific research without prior exposure to target solutions. Founded by former Google DeepMind researchers Louis Kirsch, Kaloyan Aleksiev, Tantum Collins, and Edward Hughes, the lab launched Faraday weeks after securing a $50 million seed round.

According to benchmark results published by the lab, Faraday outperformed significantly larger frontier systems, including Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5, at end-to-end scientific paper replication.

Reinforcement Learning Over Parameter Scale

Rather than training a massive proprietary foundation model from scratch, Inherent built Faraday on top of Alibaba's open-weight Qwen 3.6 27B architecture. The team applied specialized reinforcement learning algorithms to train the model on experimental design, hypothesis testing, and error correction.

Automated scientific agent workflow with reinforcement learning

Chief scientist Edward Hughes stated that the training objective focused on imbuing the agent with "research taste" (the ability to prioritize high-yield experiments and assess empirical validity) rather than memorizing methodological templates. Instead of constructing internal software tooling from scratch, Faraday delegates code generation and execution tasks to OpenAI's GPT-5.5 Codex, mirroring standard laboratory workflows where researchers leverage existing software utilities.

Benchmark Verification and Scientific Automation

Scientific paper replication serves as an established training milestone for human graduate researchers, requiring literature comprehension, methodology extraction, code reconstruction, and empirical validation. In blind evaluations where models were tasked with reproducing published results without access to source repositories or ground-truth outputs, Faraday achieved higher replication fidelity than larger general-purpose models.

Inherent operates out of King's Cross, London, employing 12 researchers with plans to expand headcount to between 20 and 25 by the end of 2026. The lab stated that paper replication represents an initial validation step toward developing generalist autonomous agents capable of formulating novel scientific hypotheses and executing discovery pipelines.

Sources

AILLMAgentsScienceresearch
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Sharpness-Aware Minimization in Large Language Models: How Adversarial Weight Perturbations and Flat Minima Boost Generalization

    In overparameterized deep neural networks, minimizing empirical training loss is insufficient to guarantee optimal generalization on unseen distributions. Modern deep architectures, including vision models and autoregressive Large Language Models (LLMs), operate in regimes where parameter counts far exceed training token counts, producing highly non-convex loss surfaces populated by infinite global minima. Standard optimization via Stochastic Gradient Descent (SGD) or AdamW often converges to sh

    1 min

  • Prefix-Tree KV Cache Management in Production: RadixAttention, Tree-Structured LRU Eviction, and Token-Level Sharing in SGLang and vLLM

    Prefix-Tree KV Cache Management in Production: RadixAttention, Tree-Structured LRU Eviction, and Token-Level Sharing in SGLang and vLLM Autoregressive large language model inference is heavily constrained by memory bandwidth and the computational overhead of the prefill phase. For workloads such as multi-turn conversations, autonomous agent tool loops, few-shot prompt evaluations, and tree-search decoding, consecutive requests often share substantial token prefixes. In a standard multi-turn ses

    1 min

  • Neural Collapse: How Simplex Equiangular Tight Frames Emerge at the Terminal Phase of Training

    In classification tasks, deep neural networks exhibit an unexpected geometric simplicity during late-stage optimization. While the internal activations of early training appear high-dimensional and complex, the penultimate layer representations and linear classifiers converge toward an exact, symmetrical geometric structure known as Neural Collapse (NC). First identified empirically by Papyan, Han, and Donoho (2020), Neural Collapse emerges during the Terminal Phase of Training (TPT). This regi

    1 min