Polish AI startup Pathway has raised additional funding at a $500 million valuation, bringing its total seed financing to $30 million as the company develops a new generation of AI models based on its post-Transformer architecture.
- Founded by Zuzanna Stamirowska and Adrian Kosowski, Pathway is developing AI models designed to improve reasoning, memory, and adaptation while reducing the computing costs associated with traditional Transformer-based systems.
- The company’s BDH (Dragon Hatchling) architecture uses persistent state, sparse activity, local interactions, and recurrent latent reasoning to process information without relying primarily on long generated reasoning chains.
- Pathway’s BDH-CQ reasoning model, with 150 million parameters, achieved 29.5% on the public ARC-AGI-1 evaluation, with a calculated inference cost of around $0.0007 per task.
- The company aims to challenge the current AI scaling approach by improving intelligence per dollar and cost-accuracy, particularly for reasoning-heavy workloads.
Details of the deal
- The new financing includes participation from Id4 Ventures, TQ Ventures, Red Bridge Ventures, Kadmos Capital, WS Investment Co., and Jonathan Frankle, Chief AI Scientist at Databricks.
- Pathway plans to use the funding to expand its compute capacity, including NVIDIA GB300 systems, and further develop and scale its post-Transformer AI architecture and BDH models.
- The company is also expanding its team, with Adam Kurzrok, former Group Product Manager for Gemini at Google DeepMind, joining as Chief Product Officer.
