River API Automates 15-Minute RL Training Following $1.1B Seed

River AI, an artificial intelligence startup founded by former xAI co-founder Igor Babuschkin, just secured $1.1 billion in early funding. The seed and Series A round, led by General Catalyst and AMP PBC, positions the two-month-old company to build infrastructure for personally trainable AI assistants. For developers, the immediate impact is the release of the River API, a backend designed to automate model fine-tuning without requiring a dedicated machine learning operations team.

The River API and Training Infrastructure

The core offering is a training backend optimized for frontier open-weight models. The platform provides direct API access for LoRA (Low-Rank Adaptation) fine-tuning and reinforcement learning. River AI specifically lists integration support for models scaling from 35 billion to 1 trillion parameters, naming Qwen3.6, Kimi K2.6, and GLM 5.2 as primary targets.

The most notable technical claim is the platform’s speed. River AI states its infrastructure can execute complex reinforcement learning training runs in 15 to 20 minutes. The system achieves this by automating model weight management, ensuring training-sampling consistency, and using elastic compute allocation.

Training costs are metered directly at $1.00 per million tokens. The company asserts this architecture makes custom model training two to four times cheaper than renting closed-source alternatives. For engineering teams currently evaluating fine-tuning vs RAG for domain-specific tasks, this cost structure lowers the barrier to entry for iterative reinforcement learning.

Hardware and Open-Weight Strategy

The capital infusion is earmarked for rebuilding the AI stack to prioritize user ownership. River AI’s roadmap moves away from centralized frontier labs, focusing instead on open-weight models that function as individual, personalized agents.

Part of the $1.1 billion will fund the development of new hardware designed to support personal AI agents running locally. This aligns with a broader industry shift toward decentralized, on-device compute environments for local AI applications, which reduce latency and keep user data out of aggregate training pools.

Ecosystem Backing

The funding round includes participation from Nvidia, AMD Ventures, Y Combinator, and Temasek. Babuschkin reportedly invested up to $100 million of his own capital into the round. While River AI has not officially confirmed its valuation, market analysts place the two-month-old company at approximately $5 billion.

General Catalyst framed the investment around maintaining U.S. leadership in the open-weight model ecosystem, categorizing the infrastructure push as a matter of national resilience.

If you build applications requiring highly specialized model behaviors, River AI’s 15-minute training latency changes the iteration cycle for reinforcement learning. You should evaluate the platform’s stability and consistent task execution times under real-world enterprise loads before migrating production workloads off of your existing training pipelines.