Inherent

Overview

Inherent is a London-based AI lab founded by ex-DeepMind researchers, focused on domain-specific scaffolding around frontier tool models for scientific research replication. Emerged from stealth on 2026-05-29 with a $50M seed round led by Index Ventures with Radical participating, and enters the corpus on 2026-08-23-AI-Digest as the first commercial ship of the harness-heavy motion this week’s research + product beats have been landing on — publishing results for its 27B research-replication agent Faraday that orchestrates GPT-5.5 Codex as a coding tool.

Timeline

2026-08-23-AI-Digest — Inherent emerged from stealth (stealth exit May 2026 at $50M Index-led seed, Radical participating) with the public disclosure of Faraday, a 27B-parameter agent purpose-built to reproduce published scientific papers end-to-end using GPT-5.5 Codex as its coding tool. Per Inherent’s own reporting, Faraday beats Claude Opus 4.8 and GPT-5.5 on the Replica paper-replication suite (310 tasks across 100 papers) at a fraction of the params. The company frames the result as evidence that domain-specific scaffolding plus tool-use design outperforms raw model size on scientific workflows. The corpus’s disciplined narrow read is that the eval is Inherent’s own suite and the “specialist scaffolding beats generalist frontier on the specialist’s own eval” framing is the historical shape of this beat and does not generalise into a “harness > weights” trend claim.

Key Developments

Stealth Exit + Faraday Ship as First Commercial Instance of the Harness-Heavy Motion (August 23, 2026): London lab founded by ex-DeepMind researchers, $50M Index-led seed (Radical participating), publishes results on Faraday — 27B agent orchestrating GPT-5.5 Codex as coding tool, reportedly beats Claude Opus 4.8 and GPT-5.5 on the Replica paper-replication suite (310 tasks / 100 papers) at a fraction of the params. Load-bearing framing to carry: specialist result on Inherent’s own eval — the correct read is “specialist scaffolding beats generalist frontier on the specialist’s own eval,” the historical shape of this beat, not a “harness > weights” generalisation. Structural read the corpus carries: Faraday lands on the same day as three of the week’s harness-and-scaffolding research beats (EnvHarness, FACET, Princeton skills paper) and two coding-agent practitioner posts — frame to carry is that harness / scaffolding / verification is where the actionable near-term work is landing this week, both in research and in shipped product; weight capability still sets the ceiling (Terminal-Bench 2.1: GPT-5.6 Sol 89.5% vs Claude Opus 5 89.1%; SWE-bench Pro: Opus 5 79.2% vs 64.6%). 30 / 60 / 90-day watch: independent third-party replication of the Replica benchmark result; whether Inherent opens the harness or the eval; whether a second scaffolding-heavy specialist ships with a similar structure inside 30 days.

Inherent

Overview

Timeline

Key Developments

Related