Goodfire Secures $150M in Series B Funding for AI Model Interpretability Tools

There is a moment in every technology cycle where the noise drops, the room gets quiet, and the adults lean forward. Not because the demo dazzles, not because the valuation flexes, but because someone finally asks the question nobody else wanted to touch. What is this thing actually doing under the hood? That is where Goodfire lives. And today, that conviction just got capitalized with a $150M Series B at a $1.25B valuation.

Goodfire is not chasing vibes or novelty. This is an AI interpretability lab built to look neural networks in the eye and ask them to explain themselves. Founded in June 2024 in San Francisco, the company is a Public Benefit Corporation by design, which already tells you this is not a fast money tourist operation. Co founder and CEO Eric Ho, alongside Co founder and CTO Daniel Balsam, and Co founder and Chief Scientist Tom McGrath, PhD, came together around a shared belief that black box models are not a feature, they are a liability.

Eric Ho and Daniel Balsam have history. They built together before, scaled real revenue, and walked away when the problem stopped being interesting. Tom McGrath, PhD brought the scars and signal from building interpretability at Google DeepMind. When those threads tied, Goodfire became a lab obsessed with mechanistic interpretability, the unglamorous but essential work of understanding why models behave the way they do, neuron by neuron, circuit by circuit.

The flagship platform, Ember, does something most teams talk around and few can do. It decodes what is happening inside models and gives developers programmable access to those internals. Not prompts. Not guardrails. Actual steering. That is why Goodfire was able to interpret DeepSeek R1 when nobody else could. That is why Arc Institute trusted Goodfire to work on the Evo 2 DNA foundation model. Fire, but controlled. Heat with intent.

This Series B was led by B Capital, with returning conviction from Juniper Ventures, Menlo Ventures, Lightspeed Venture Partners, South Park Commons, and Wing Venture Capital. New belief showed up from DFJ Growth, Salesforce Ventures, and Eric Schmidt. Capital does not align like that unless the room agrees the problem is real and the approach is rare.

The lesson here is not that AI safety suddenly became fashionable. It is that rigor compounds. Goodfire graduated from the Anthology Fund into a full Series A, then kept proving that interpretability could be commercially viable without watering it down. When you can open the model, understand it, and shape it, you do not just reduce risk. You unlock performance, discovery, and trust.

Goodfire is hiring, building, and pushing toward intentional design of models that can be understood, not guessed at. In a market addicted to speed, this team chose clarity. That choice just set a lot of future agendas, and the rest of the ecosystem is going to have to decide whether it wants to keep guessing or finally learn how the fire actually works.