/Tech7h ago

River AI launches River API, enabling post-training and reinforcement learning on open-source models up to 1 trillion parameters

The startup is backed by xAI co-founder Igor Babuschkin.

3157747208162.5K
Banghua Zhu
Beff (e/acc)
Igor Babuschkin#128
Original post
Igor Babuschkin@ibab#128inTech

We are releasing River API, our first product, in early access. The API gives you access to the same battle-tested tools that we’re using internally at River for post-training, reinforcement learning and continual learning. Check it out and let us know what you think!

River AI@river_ai_inc

Introducing River API. Fine-tune and RL train leading open-source models at scale, ranging from 35B to 1T params. We’ve been using it internally to power our research and we love it. Today, we are opening up our public waitlist. Own your intelligence!

2:17 PM · Jun 10, 2026 · 75.8K Views
Sentiment

Users are excited about River's new API for fine-tuning and RL training of 1T-parameter open-source models because it matches their interests and they've already signed up for the public waitlist.

Pos
100.0%
Neg
0.0%
6 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS950BOOKMARKS1
River AI@river_ai_inc

Sign up here: https://river.ai/api

7hViews 950Likes 4Bookmarks 1
LIKES5
Beff (e/acc)@beffjezos

@ibab this is very cool

Igor Babuschkin@ibab

We are releasing River API, our first product, in early access. The API gives you access to the same battle-tested tools that we’re using internally at River for post-training, reinforcement learning and continual learning. Check it out and let us know what you think!

4hViews 414Likes 5Bookmarks 0
RETWEETS11
River AI@river_ai_inc

Introducing River API. Fine-tune and RL train leading open-source models at scale, ranging from 35B to 1T params. We’ve been using it internally to power our research and we love it. Today, we are opening up our public waitlist. Own your intelligence!

7hViews 91.3KLikes 190Bookmarks 77
Nikolai Yakovenko@ivan_bezdomny

@ibab Nice. I’ve been RL finetuning open source models and will give this a shot!

7hViews 139Likes 2
Ely Hahami@ElyHahami

@ibab how is this different from eg tinker?

7hViews 103Likes 2
Zion@BlasianHokage

Congratulations Igor and very aligned mission. I imagine there will be updates soon for more user friendly versions for more consumer use cases soon. In which case the data portability to power the post, rl and cl will be very important

Have you look at @onairosapp for data portability?

7hViews 73Likes 1
Saylor@seylorra

@ibab already submitted to the waitlist

curious how much of the 1T setup they let users actually touch in early access

6hViews 191
Benjamin Mowinckel@benmowinckel

@ibab Personal data collection needs private personal systems with agnostic local storage no? Would love to hear your thoughts on whats best for how we all navigate these issues - im trying my best to solve it at Alexandria!

6hViews 63
Geet Khosla@geetkhosla

@river_ai_inc signed up, can’t wait to test it out.

1hViews 47
neekhil vatsa@garfieldII

@ibab @jack Great work, and do well on against your mission!

3hViews 40
Lovedognot idol@Lovedognotidol

Your post made me think of an old virtual pet game called Tamagotchi.

When a Tamagotchi is first born, it isn't good or bad. It has a basic structure and a set of rules, but how it develops depends largely on how its owner raises it.

River AI feels similar to me. If users can continuously train and shape their own AI, then I wonder:

What is the AI's original nature before any personalization takes place?

Is its core identity simply pattern recognition and prediction, or does it already inherit certain values, assumptions, and objectives from its creators?

And if users keep training it over time, can its original logic truly change, or are we only modifying its behavior and preferences while the underlying foundation remains the same?

In other words, are we raising the AI, or are we revealing what was already hidden inside it from the beginning?

I'd love to hear your thoughts.

1hViews 33
Nazanin@nazaninzah22065

@ibab Ohom 🤔

6hViews 32
REAZ@LLMathematician

@ibab Okayi

6hViews 29
matthias@mmatthias

@river_ai_inc Does it have tool call?

6hViews 22
Tanmay@tanny2109

@river_ai_inc How does this compare to tinker?

1hViews 12
Lovedognot idol@Lovedognotidol

Maybe my real question is this:

If River allows users to shape and personalize their AI over time, what remains unchanged underneath all that learning?

In other words, what is River AI's equivalent of "genetics" before the user's training begins?

When two users train their models in completely different directions, is there still a shared core identity inherited from the foundation model, or can that core itself evolve?

I'm less curious about what users can teach the AI, and more curious about what the AI already is before anyone teaches it.

54mViews 7
mirajul@Mirajulctn

@ibab @jack 🤖 AI & ChatGPT Guides https://mafijul.gumroad.com/l/ryldc https://mafijul.gumroad.com/l/ntyujo https://mafijul.gumroad.com/l/ooiwzl https://mafijul.gumroad.com/l/yktwe https://mafijul.gumroad.com/l/szxmzh https://mafijul.gumroad.com/l/jikdlt https://mafijul.gumroad.com/l/suipqp https://mafijul.gumroad.com/l/flrpy

1h
mirajul@Mirajulctn

@ibab @jack 🤖 AI & ChatGPT Guides https://mafijul.gumroad.com/l/ryldc https://mafijul.gumroad.com/l/ntyujo https://mafijul.gumroad.com/l/ooiwzl https://mafijul.gumroad.com/l/yktwe https://mafijul.gumroad.com/l/szxmzh https://mafijul.gumroad.com/l/jikdlt https://mafijul.gumroad.com/l/suipqp https://mafijul.gumroad.com/l/flrpy

2h
Blissy@BlissyOnX

@ibab rolling the same tools out publicly is always the real signal

whats the one thing you built with it internally that surprised u most?

7h
Load more posts