MaineCoon vs Goodfire

The two feature lists barely overlap in intent. MaineCoon's stated work is building a real-time audio-visual social world model — perceiving social interaction across both modalities simultaneously, with architecture and methodology documented in technical reports. Its output is knowledge, not software you license; the description explicitly notes no public product, API, or pricing. Goodfire's Silico is the opposite: a productized interpretability stack with a defined Understand/Debug/Design loop. Understand exposes learned internal structure; Debug predicts which RL preference data will amplify or suppress a behavior, amplifies checkpoint differences in logit space to surface unexpected behaviors, detects performative chain-of-thought for early exit, and traces rare unstable behaviors in robotics policies via latent structure. Design steers models — features-as-training-rewards cut hallucinations by 58%. Breadth is unusual: block-sparse featurizers on vision models, activation harvesting from trillion-parameter models, interpreting weights (not just activations), explaining 4.2 million ClinVar genetic variants, and surfacing Alzheimer's biomarkers from epigenetic models. MaineCoon's listed audience is researchers and world-model developers; Goodfire's is research teams with a debugging, validation, or control problem. If you need to ship a multimodal social-perception system today, neither hands you a turnkey API — but only Goodfire hands you tooling at all.

The pricing models aren't comparable in kind. MaineCoon is tagged free, and that reflects reality: it's a research project publishing technical reports and project links, with no publicly available products, APIs, or pricing as of its latest update. That's not a free tier — it's a pre-commercial artifact. There's no per-seat cost, no usage meter, nothing to cancel. Goodfire is freemium, so there's a free entry path into Silico with paid tiers above it for teams running it in anger; the public data here doesn't list tier prices, seat minimums, or usage limits, so don't budget off a number you saw elsewhere. The practical distinction: MaineCoon costs you time — reading reports, tracking an early research effort that may never productize. Goodfire costs money once you exceed the free tier, but you're buying a maintained platform whose roadmap includes a research grants program and a SOC 2 Type II certification, which is the kind of artifact procurement and security review will ask for. If budget approval is part of your process, only one of these has anything to approve.