Show HN: Sup AI, a confidence-weighted ensemble (52.15% on Humanity's Last Exam)
14 hours ago
- #Transparency
- #AI Accuracy
- #Ensemble Learning
- Sup AI is the most accurate AI in existence, featuring 337 models and real-time logprob scoring.
- Humanity's Last Exam (HLE) is a challenging benchmark with 2,500 questions across 100+ subjects, created by domain experts.
- Sup AI achieves 52.15% accuracy on HLE, 7+ points ahead of individual models in its ensemble.
- The AI uses real-time logprob confidence scoring and cross-model disagreement detection to verify answers.
- Sup AI employs ensemble search across multiple retrieval methods for thorough document and file intelligence.
- The platform supports 10 GB uploads and features lossless context compaction for infinite context handling.
- Sup AI provides complete transparency with visible sources, document citations, and file references.
- The model ecosystem includes 337 models from 50+ providers, with intelligent orchestration for optimal performance.
- Cost-optimized ensemble ensures better answers at nearly the same price as running a single model.
- Free credits are available for trying all AI models with full feature access.