Features
OpenAI-compatible chat completions endpoint
Calibrated probabilistic forecasts (0-1 scores)
Auto-research context gathering
Multiple model versions (v3, v4 Low, v4 Full)
Live prediction market integration
Custom model training on domain-specific data
Future-as-Label reinforcement learning training method
Benchmark-verified accuracy on ProphetArena and ForecastBench
Support for supply chain, clinical, SEC filing risk forecasting
Brier Skill Score up to 26% (v4 Full)
Cheaper inference than GPT-5, Gemini, Opus
Enterprise-grade API with rate limits and security
Clinical event forecasting from raw notes (research)
Domain-specific SLLM training on customer data
On-premises or dedicated EU-cloud deployment
Compliance with GDPR and EU AI Act
Collaborative evaluation tools for client teams
PhariaFiles for enterprise file processing
Tokenizer-free model TFree-HAT 7B
Alpha-MoE megakernel for faster tensor-parallel inference
Theoretical inference model for DeepSeek hardware
Secure handling of sensitive documents (90% faster search)
40% faster RFQ processing via AI agents
End-to-end support from ideation to deployment
Integration with internal knowledge worker workflows
