Closed-source AI systems often lack adaptability and agility. They can be rigid, evading simple questions, injecting unwanted moralizing, or demanding intricate prompts to produce anything worthwhile. Businesses struggle to customize these models, regardless of how ingenious their prompts or sleek their interfaces. This results in a draining and intensive cycle of prompt refinement that is unsustainable and uninspiring.

At River, we are driven to create AI that feels vibrant, seamlessly syncing with individual aspirations and imaginative visions. Introducing River 1b—our answer to the personalization gap.

River 1b is open-source, unfiltered, and highly steerable. It was crafted by fine-tuning Llama 3.1 (8B, 70B, and 405B) on a dataset largely composed of synthetic responses. This approach allows River 1b to match or surpass Llama 3.1’s performance while delivering richer reasoning and creative depth. River 1b excels at long-context retention, multi-turn conversations, intricate roleplaying, internal monologues, and agentic functions on platforms like x, Discord, Telegram, and Meta.

Our training approach ensures that River 1b follows system and instruction prompts with client-chosen precision and versatility, making it a tool that truly resonates with brands white labeling the product.