River AI

River AI

River AI is a training and serving platform for teams that want to own the model rather than rent a prompt. River AI exposes an API and a Python client for fine-tuning open models from roughly 35B to 1T parameters, supporting both LoRA adaptation and reinforcement learning, then serves the result back through the same API. The company's framing — bring your data, shape the weights, own the intelligence you create — puts it against the inference clouds that host someone else's models. River AI was founded by xAI co-founder Igor Babuschkin and raised at a scale rarely seen this early.

Production credibility: River AI announced $1.1B across a seed and Series A on 11 August 2026, led by General Catalyst and AMP PBC, when the company was roughly two months old — one of the largest early rounds on record. It was founded by Igor Babuschkin, a co-founder of xAI. The product is live rather than announced: the site publishes an API billed per million tokens, with rates depending on the open model chosen, plus LoRA fine-tuning and reinforcement learning through a Python client.

Key Features

• Fine-tuning across open models from roughly 35B to 1T parameters
• LoRA adaptation and reinforcement learning from the same client
• Serving through an API once a model has been trained
• Python client for the training and deployment loop
• Pay-per-token billing, priced by the underlying open model
• Customer-owned weights rather than prompts against a hosted model

Ideal Use Case

Teams whose product depends on model behaviour they cannot get from prompting: domain-specific assistants, agents that must follow house conventions, or workloads where a smaller tuned model beats a larger general one on cost. It also fits organisations with a data asset they want reflected in weights they control rather than in a context window.

How River AI differentiates

Together, Fireworks, Baseten and Replicate mostly host models and sell inference; fine-tuning is a feature alongside it. River AI inverts that: training is the product, and serving is what happens afterwards. Reinforcement learning as a first-class API, not just supervised fine-tuning, is the sharper technical difference. The open question is maturity — the company is months old, so the platform is young next to inference providers with years of production traffic, and the funding, not the track record, is what makes it worth watching.

FAQ

Q: What does River AI do? A: River AI lets teams train, tune and serve their own models. It supports LoRA fine-tuning and reinforcement learning across open models from about 35B to 1T parameters, then serves the result through its API.

Q: How much does River AI cost? A: River AI bills per token, with rates depending on the open model used. The site does not publish a flat subscription price.

Q: Who founded River AI? A: River AI was founded by Igor Babuschkin, a co-founder of xAI.

Q: How much has River AI raised? A: River AI announced $1.1B across a seed and Series A on 11 August 2026, led by General Catalyst and AMP PBC, when the company was around two months old.

Q: How is River AI different from Together AI? A: Together and similar providers primarily host and serve existing models. River AI is built around training first — LoRA and reinforcement learning on open weights — with serving as the step that follows.

tl;dr

River AI is a platform for training and serving your own open models, with LoRA and reinforcement learning behind a pay-per-token API. Founded by xAI co-founder Igor Babuschkin, it raised $1.1B across seed and Series A in August 2026, led by General Catalyst and AMP PBC.