Goodfire Hits $1.25 Billion Valuation With $150 Million Series B to Scale Interpretability-Centric AI Platform

New updates have been reported about Goodfire.

Goodfire has raised $150 million in a Series B round at a $1.25 billion valuation, less than a year after its Series A, positioning the San Francisco-based AI interpretability lab as a newly minted unicorn. The round was led by B Capital with participation from existing backers Juniper Ventures, Menlo Ventures, Lightspeed Venture Partners, South Park Commons, Wing Venture Capital, and new investors including DFJ Growth, Salesforce Ventures, and Eric Schmidt. Capital will be deployed to advance Goodfire’s frontier research, build the next generation of its interpretability-driven “model design environment,” and scale commercial and research partnerships, particularly in AI agents and life sciences. Goodfire’s approach centers on reverse-engineering and directly editing internal neural network mechanisms, enabling model steering, safety improvements, and AI-to-human knowledge transfer; the company has already used these techniques to identify a novel class of Alzheimer’s biomarkers from an epigenetic foundation model built by Prima Mente, in collaboration with partners such as Mayo Clinic and Arc Institute.

On the model design side, Goodfire has developed methods to retrain and adjust behavior by targeting specific subcomponents inside large models, reporting a 50% reduction in hallucinations in a large language model via precise internal interventions rather than broad retraining. The new funding will support productizing this capability into a scalable platform that allows enterprises to inspect, debug, and modify model behavior with software-like control, aiming to shift AI development away from opaque “black box” training toward interpretable, controllable systems. Goodfire’s CEO Eric Ho frames interpretability as the enabling science for “designing intelligence,” analogous to thermodynamics for engines, and the company positions itself as a research-first “neolab” pursuing training and model-understanding breakthroughs overlooked by major scaling-focused labs. The team includes leading interpretability researchers from DeepMind, OpenAI, and top universities, such as Nick Cammarata, co-founder Tom McGrath, and UC San Diego professor Leon Bergen (on leave). Strategically, the raise gives Goodfire substantial runway to deepen its scientific discovery pipeline, expand into high-value domains where models already surpass human intuition, and commercialize tools that could materially influence how enterprises build, govern, and extract value from advanced AI systems.