Goodfire Raises $150 Million at $1.25 Billion Valuation to Decode AI Models and Improve Safety

Goodfire Secures $150 Million in Series B Funding at $1.25 Billion Valuation

Goodfire has closed a $150 million in Series B funding round led by B Capital, with participation from Menlo Ventures and Lightspeed Venture Partners, along with contributions from Salesforce and former Google CEO Eric Schmidt2

2

. The deal values the San Francisco-based startup at $1.25 billion valuation, bringing its total raised capital to $209 million since its founding in 20241

1

4

4

. The fresh capital will fund the startup's mission to decode AI models and make their decision-making processes transparent, addressing what CEO Eric Ho calls a "reckless" approach to AI development.

Understanding How AI Models Make Decisions Through Interpretability

Goodfire specializes in AI interpretability, the science of reverse engineering neural networks to understand what AI models are actually doing when they process information3

3

. The challenge stems from the fundamental architecture of large language models (LLMs), which consist of artificial neurons that interact in extraordinarily complex ways—tens of thousands of neurons can be involved in generating a single prompt response2

2

. The weights underlying an AI model's performance are written in code that's indecipherable to humans, making it nearly impossible for developers to debug problems without completely retraining models1

1

.

Model Design Environment Enables AI Model Understanding

To break through this opacity, Goodfire has built what it calls a model design environment—a platform that uses AI interpretability methods to map out the internal components of foundation models2

2

3

3

. The company creates interpreter models that effectively map the mind of an AI model, then perform what Ho describes as "brain surgery" to improve performance or derive novel insights1

1

. The platform operates across two critical phases: during training, it maps out the learning workflow and identifies flaws to boost output quality, and in production, it monitors model performance to catch issues like AI hallucinations, which Goodfire claims to have reduced by half in one recent project2

2

.

Real-World Impact: Discovering Alzheimer's Biomarkers

The practical applications of Goodfire's technology extend into critical healthcare domains. Prima Mente, a healthcare AI startup, developed an AI model to detect Alzheimer's disease by analyzing cfDNA fragments, but the team didn't fully understand how the model made its predictions1

1

. Using Goodfire's interpreter technology, researchers discovered a novel class of Alzheimer's biomarkers based on the length of cfDNA fragments—connections the model was making that humans had not previously identified in existing scientific literature2

2

. This breakthrough demonstrates how debugging AI can unlock scientific insights hidden within neural networks.

Growing Client Base and Technical Innovation

Goodfire works with major clients including Microsoft, the Mayo Clinic, and the nonprofit Arc Institute4

4

. Some organizations use the platform to build their own AI tools using existing foundation models, while model makers leverage it to enhance AI performance. Last year, the company developed a method called SPD that identifies which model components are involved in generating prompt responses by systematically removing components and observing the effects on output2

2

. Eric Ho frames this work as essential: "Interpretability, for us, is the toolset for a new domain of science: a way to form hypotheses, run experiments, and ultimately design intelligence rather than stumbling into it"2

2

.

Investment Plans and Industry Context

With the new funding, Goodfire plans to advance from debugging existing models to retraining AI models for better performance, while also investing heavily in computing power and expanding its team1

1

4

4

. The startup joins a growing class of AI research companies sometimes called neolabs that have commanded significant valuations. OpenAI alums Mira Murati and Ilya Sutskever have secured billions for their ventures, while AI researcher Richard Socher is in funding talks at a $4 billion valuation1

1

. Ho's concerns about the current trajectory remain stark: "I don't like the trajectory of AI right now—we're about to deploy all these systems that we don't understand everywhere. I think that's not good, and I want to change that"1

1

.