Hey HN, we’re Andrew and Derik at RunRL (https://runrl.com/). We've built a platform to improve models and agents with reinforcement learning. If you can define a metric, we'll make your model or agent better, without you having to think about managing GPU clusters.

Here's a demo video: https://youtu.be/EtiBjs4jfCg

I (Andrew) was doing a PhD in reinforcement learning on language models, and everyone kept...not using RL because it was too hard to get running. At some point I realized that someone's got to sit down and actually write a good platform for running RL experiments.

Once this happened, people started using it for antiviral design, formal verification, browser agents, and a bunch of other cool applications, so we decided to make a startup out of it.

How it works:

- Choose an open-weight base model (weights are necessary for RL updates; Qwen3-4B-Instruct-2507 is a good starting point)

- Upload a set of initial prompts ("Generate an antiviral targeting Sars-CoV-2 protease", "Prove this theorem", "What's the average summer high in Windhoek?")

- Define a reward function, using Python, an LLM-as-a-judge, or both

- For complex settings, you can define an entire multi-turn environment

- Watch the reward go up!

For most well-defined problems, a small open model + RunRL outperforms frontier models. (For instance, we've seen Qwen-3B do better than Claude 4.1 Opus on antiviral design.) This is because LLM intelligence is notoriously "spiky"; often models are decent-but-not-great at common-sense knowledge, are randomly good at a few domains, but make mistakes on lots of other tasks. RunRL creates spikes precisely on the tasks where you need them.

Pricing: $80/node-hour. Most models up to 14B parameters fit on one node (0.6-1.2 TB of VRAM). We do full fine-tuning, at the cost of parameter-efficiency (with RL, people seem to care a lot about the last few percent gains in e.g. agent reliability).

Next up: continuous learning; tool use. Tool use is currently in private beta, which you can join here: https://forms.gle/D2mSmeQDVCDraPQg8

We'd love to hear any thoughts, questions, or positive or negative reinforcement!

Launch HN: RunRL (YC X25) – Reinforcement learning as a service

Granite docling 258M: a small multimodal model for efficient document conversion

Proximity screening pushes graphene electronic quality to record levels

Trump administration retreats on combating human trafficking, child exploitation

RetroPilot V1 – Teaser [video]

Gnome 49 Release

Planned Cities

Next.js GET Cache Bypass via Multipart POST – Reproduces on Vercel.com

Galactica, the doomed model launched two weeks before ChatGPT

The State of AI and Tech Q2 2025 Industry Report

$599 MacBook with iPhone Chip Expected to Enter Production This Year

Ask HN: Is Claude Code less useful in recent weeks for you?

Online therapy for expats and digital nomads

Show HN: Xiaoniao – Paste-as-Translation (Go and AI)

Not Buying American Anymore

Startup Working to Bring Back Dodo Bird Raises $120M

Microbial iron oxide respiration coupled to sulfide oxidation

Giant Subterranean Neutrino Detector Is Taking on the Mysteries of Physics

A Fusion-Reactor-Inspired Thruster Could Deorbit Space Junk

Generating Blue Noise Sample Points with Mitchell's Best Candidate Algorithm

What It's Like to Work Inside a Broken CDC

Frivolous, unethical and unjustifiable- SF agency misspent $4.6M audit finds

Tell HN: Discord is apparently rolling out age verification for EU/EEA residents

Did NASA's Perseverance rover find evidence of ancient life on Mars?

Screen readers do not need to be saved by AI

Moving off of TypeScript, 2.5M lines of code

A Guide to Midnight Commander (2012)

Show HN: I did a 4 hour conversational audiobook on the history of data centers

Gmail Mail Delivery Subsystem Being Used for Spam Delivery Bypassing Filters

What can we learn from Spotify layoffs? (2024)

Ask HN: Ask HN page this week restricted to a small list