Goodfire

Goodfire develops tools and research to understand how AI models work internally and change their behavior. Its Silico interpretability agent helps AI teams investigate learned features, debug unexpected predictions, and replicate research on their own models and datasets. Given a research goal, Silico plans experiments, runs work across computing nodes, monitors progress, and returns inspectable results. Its tools include probes, sparse autoencoders, causal analysis, and data attribution, with applications spanning language models, life sciences, robotics, and vision.

Founded in 2024, Goodfire is led by co-founders Eric Ho, CEO; Daniel Balsam, CTO; and Tom McGrath, Chief Scientist. Its research connects model inspection with training: Reinforcement Learning from Feature Rewards (RLFR) uses probes of internal representations to supply reinforcement-learning rewards, turning signals inside a model into feedback for changing its behavior. In genomics, its collaboration with Mayo Clinic produced EVEE, a freely accessible resource using Evo 2 representations to provide predictions and mechanistic hypotheses for 4.2 million ClinVar variants—research outputs, not diagnoses.

Goodfire’s scientific-model partnerships include Mayo Clinic, Arc Institute, and Prima Mente. In 2026, the company announced a $150 million Series B led by B Capital at a $1.25 billion valuation, bringing its reported cumulative backing above $200 million. The funding supports interpretability research, model-design tools, and expansion of partnerships across AI agents and life sciences.