River AI
River AI is building personal intelligence and an API for fine-tuning open models. It targets teams that want to own and serve the models they create.
About River AI
River AI is building personal intelligence that people can own and shape, spanning interfaces, models, algorithms, infrastructure, and personal hardware. Its current API gives researchers and engineering teams a way to fine-tune open-source models with LoRA and reinforcement learning, then sample and serve the resulting checkpoints.
The API is designed around bringing your own data and reward signals to a model-training workflow. River supports open models across several sizes and context lengths, provides a small Python client, and treats the trained checkpoint as an asset that the customer can keep, iterate on, and deploy.
River’s API preview uses token-based billing for inference and training. The public rate table currently starts at $0.33 per 1 million prompt tokens for Qwen3.6-35B-A3B-FP8, with completion, training, cached-token, and checkpoint-storage charges varying by model and usage.
Discover River AI and thousands of other AI tools on Beyond The AI, your trusted directory for AI solutions.
Who is using River AI?
Key Features
- LoRA-based fine-tuning
- Reinforcement learning on custom rewards
- Python client for training and inference
- Open model support from 35B to 1T parameters
- Multiple context-length options
- Token-based usage billing
- Checkpoint ownership and serving
Use Cases
- Fine-tune open models for specialized tasks
- Train models against internal reward signals
- Build cost-efficient task-specific AI agents
- Evaluate and iterate on custom model checkpoints
- Serve trained models through production endpoints
