Forges
Models, trained by whoever shows up.
A Forge is a collective that trains one open model — an Ingot. Anyone can add data, compute, or capital. A TEE eval measures how much each contribution actually moved the model, and ownership mints in exactly that proportion. Revenue from running the Ingot routes back to owners on-chain — forever.
Contribute
Bring openly licensed data, run training/eval compute, or fund the eval set. Every input is logged on 0G.
Get measured
The eval coordinator scores each contribution's marginal effect on a held-out set inside a trusted enclave.
Own & earn
Ingot shares mint proportionally. Inference revenue then splits to contributors automatically.
Classification
LIVEFinancial Filing Sentiment Classifier
A classifier that scores sentiment and risk tone in 10-K / 10-Q filing sections, tuned for the hedged language of financial disclosure.
Generation
OPENRust Unsafe-Code Auditor
A code model that reviews Rust `unsafe` blocks and explains the soundness obligations a reviewer must check.
Translation
LIVEYorùbá ↔ English Translator
An open translation model for Yorùbá, a tonal West-African language with ~45M speakers and no production-grade MT.
Embedding
LIVEMultilingual Legal Citation Embeddings
A dense embedding model for retrieving relevant case law across jurisdictions, where generic embeddings miss legal-citation semantics.
Classification
LIVEICD-10 Medical Coding Classifier
A classifier that maps free-text clinical encounter notes to ICD-10 diagnosis codes, automating a slow, error-prone billing step.
Generation
LIVEBengali News Summarizer
An abstractive summarizer for Bengali news, a language with 270M+ speakers but almost no production summarization models.
Classification
LIVESpoken-Hindi Intent Classifier
An intent classifier for code-mixed spoken Hindi-English (Hinglish), built for voice interfaces that India's existing NLU stacks handle poorly.
Generation
LIVESolidity Vulnerability Explainer
A code model that reads a Solidity function and explains likely vulnerability classes in plain English — reentrancy, unchecked math, access-control gaps.
Classification
LIVEClinical Note De-Identification
A model that detects and redacts protected health information in free-text clinical notes, so hospitals can share data for research without leaking patient identity.
Embedding
OPENChemistry-Paper Retrieval Embeddings
A dense embedding model tuned for semantic retrieval over chemistry and materials-science literature, where general-purpose embeddings miss reaction and compound nuance.
Classification
LIVEMSA Risk-Clause Classifier
A contract-intelligence model that flags risky clauses in software Master Service Agreements — uncapped liability, auto-renewal, IP assignment, unilateral termination.
Translation
LIVEKonkani ↔ English Translator
An open-weights translation model for Konkani, a low-resource Indian language with no production-grade MT. Trained from community-contributed parallel text.