Pathway

Pathway is a frontier AI lab building two things: BDH (Dragon Hatchling), a post-transformer AI architecture that unifies memory and reasoning, and the Live Data Framework, an open-core engine for real-time AI pipelines. The company recently hit a $500 million valuation as it prepares to release public benchmarks for its post-transformer models, and it claims its 150M-parameter model achieves 29.5% Pass@2 on ARC-AGI-1 at $0.0007 per task — an 11x lower reasoning cost than existing models. BDH also reports 97.4% solve rate on Sudoku Extreme without chain-of-thought and 95% accuracy on BABILong at 32K context, though these results are pending final contamination checks and independent validation. For data engineers and ML teams, the Live Data Framework is the more immediately usable piece. It's a Rust-based true streaming engine that handles batch and streaming workloads with the same code, offering incremental stateful operations, built-in HNSW vector search plus hybrid search, and a REST API with sub-millisecond latency. You can write pipelines in Python or SQL, run them in Jupyter notebooks, and connect to 300+ data sources via predefined connectors. The Community tier is free under BSL 1.1 and self-hosted; Scale adds monitoring and advanced connectors; Enterprise brings horizontal scalability, high availability, and managed services. Pathway positions itself as an alternative to Flink for streaming infrastructure, but with a sharper focus on AI workloads like live RAG, streaming ETL, and real-time feature serving. BDH is the flashier bet — potentially cheaper reasoning with a brain-inspired architecture — but it's still on waitlist for production use. For teams that want a self-hosted, open-core engine today, Pathway delivers; for those betting on post-transformer models, the public benchmarks are the moment to watch.