Top all-time · #5580
OpenPipe Blog
Company blog · openpipe.ai/blog · on HN as kcorbittaiprogramming
OpenPipe Blog is the company blog at openpipe.ai, writing about ai and programming. Since 2023, 14 different Hacker News readers have linked it 39 times, most often “Fine-Tuning for Production Apps”. It is still writing: the last post was on 2026-05-18, and you can follow it by feed.
Too few sampled comments (4) to say how often HN recommends it.?
Why HN links it
HN readers link OpenPipe's blog for posts on fine-tuning models to beat GPT-4 and a mixture-of-agents cost-versus-performance comparison. One commenter mentions the ART reinforcement-learning library.
Summarised by Claude from a sample of 4 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
The OpenPipe platform has migrated to Weights & Biases and CoreWeave. Browse migration details and the OpenPipe technical archive.
Most-linked posts?
- Fine-Tuning for Production Apps4 people · 6 mentions · first linked 2024-04-24
- ART·E: how we built an email research agent that beats o32 people · 4 mentions · first linked 2025-04-29
- OpenPipe Mixture of Agents: Outperform GPT-4 at 1/25th the Cost2 people · 3 mentions · first linked 2024-06-20
- Mistral 7B Fine-Tune Optimized2 people · 2 mentions · first linked 2023-12-20
- Fine-Tuning Best Practices Series Introduction and Chapter 1: Training Data2 people · 2 mentions · first linked 2024-08-02
- LLM Fine-Tuning Best Practices: Base Models Proprietary/Open Source, Large/Small2 people · 2 mentions · first linked 2024-08-28
- Using reinforcement learning and $4.80 of GPU time to find the best HN post2 people · 2 mentions · first linked 2024-10-28
What Hacker News says?
This actually already exists! We did a writeup of the relevant optimizations here: https://openpipe.ai/blog/s-lora
Comment linking /blog/s-lora
on “Groq runs Mixtral 8x7B-32k with 500 T/s”Submitted Using reinforcement learning and $4.80 of GPU time to find the best HN post as a story · 217 points
HN discussionI actually did a blog post a few months ago where I analyzed HN commenter sentiment across AI, blockchain, remote work and Rust. The final graph at the very end of the post is the relevant one on this topic! https://openpipe.ai/blog/hn-ai-crypto
Comment linking /blog/hn-ai-crypto
on “Show HN: Exploring HN by mapping and analyzing 40M posts and comments for fun”Submitted Using GRPO to Beat o1, o3-mini and R1 at “Temporal Clue” as a story · 199 points
HN discussionOpenPipe | ML & Full-Stack Engineers | Full-time | Seattle, WA (ONSITE) | https://openpipe.ai/ | Highly competitive pay + equity We've built the world's best fine-tuning platform in just over a year. First to launch self-service preference tuning, integrated evals/data prep/fine-tuning, and self-service learning from human feedback. Our customers range from fast-growing startups to Fortune 500s. We're growing 40% Mo…
Comment linking /
on “Ask HN: Who is hiring? (February 2025)”Submitted Show HN: RULER – Easily apply RL to any agent as a story · 81 points
HN discussionFor "that last 10% of reliability" RL is actually working pretty well right now too! https://openpipe.ai/blog/art-e-mail-agent
Comment linking /blog/art-e-mail-agent
on “The unreasonable effectiveness of an LLM agent loop with tool use”Submitted OpenPipe Mixture of Agents: Outperform GPT-4 at 1/25th the Cost as a story · 13 points
HN discussion(Disclaimer: I'm the founder of OpenPipe, one of the fine-tuning services OP tried and ultimately the one that produced the highest performing model, it appears.) Data extraction is a use case that fine-tuned models are fantastic at, so I'm not surprised that OP got good results. That said, I've also found it's pretty easy to beat GPT-4 across many task types if you have a way of getting strong training data. We pub…
Comment linking /blog/mixture-of-agents
on “My finetuned models beat OpenAI's GPT-4”Submitted Serverless RL: Faster, Cheaper and More Flexible RL Training as a story · 9 points
HN discussionOpenpPipe - https://openpipe.ai/ - is probably the service that most closely resembles what you’re asking for, but I found the evals weren’t really what I wanted — i.e. following my custom evaluation criteria — so you probably will end up having to do that yourself anyway. But for the finetuning, they’re all somewhat the same. Predibase and OpenPipe are two good options for that. Predibase has more base models for y…
Comment linking /
on “My finetuned models beat OpenAI's GPT-4”Submitted PII-Redact – SOTA PII Redaction on Your Laptop as a story · 6 points
HN discussionOpenPipe | ML & Full-Stack Engineers | Full-time | Seattle, WA (ONSITE) | https://openpipe.ai/ | Highly competitive pay + equity We've built the world's best fine-tuning platform in just over a year. First to launch self-service preference tuning, integrated evals/data prep/fine-tuning, and self-service learning from human feedback. Our customers range from fast-growing startups to Fortune 500s. We're growing 40% Mo…
Comment linking /
on “Ask HN: Who is hiring? (January 2025)”Submitted Is AI the Next Crypto? Insights from 2M HN Comments as a story · 6 points
HN discussionHah... Ask, and HN (sorta) delivers. Not RNN, but self-finetuned LLM cost vs performance compared to GPT-4. https://openpipe.ai/blog/mixture-of-agents
Comment linking /blog/mixture-of-agents
on “My Python code is a neural network”Submitted Analyzing OpenAI's Reinforcement Fine-Tuning: Less Data, Better Results as a story · 4 points
HN discussion<< Higher ups may want it done, but probably only if it can be done immediately and doesn't cost a lot of money. Since we are talking evil, I suppose money is a good place to start ( being root of it and all that ). So motivation would likely be there, but I am willing to accept your qualification of 'yes, but don't spend a lot'. Lets look at some of that potential cost structure. The analytics part these days is no…
Comment linking /blog/hacker-news-rlhf-part-1
on “The sins of the 90s: Questioning a puzzling claim about mass surveillance”Submitted Everything I know about reward hacking as a story · 3 points
HN discussionObviously talking my own book here, but we've helped dozens of customers make the transition from prompted GPT-4 or GPT-3.5 to their own fine-tuned models at OpenPipe. The most common reaction I get is "wow, I didn't expect that to work so well with so little effort". For most tasks, a fine-tuned Mistral 7B will consistently outperform GPT-3.5 at a fraction of the cost, and for some use cases will even match or outp…
Comment linking /blog/mistral-7b-fine-tune-optimized
on “Ask HN: People who switched from GPT to their own models. How was it?”Submitted ART·E: how we built an email research agent that beats o3 as a story · 3 points
HN discussionIMO it's possible to over-generalize from this datapoint (lol). While it's true that creating a general "finance" model that's stronger than GPT-4 is hard, training a task-specific model is much easier. Eg. "a model that's better than GPT-4 at answering finance-related questions": very hard. "A model that's better than GPT-4 at extracting forward-looking financial projections in a standard format": very easy. And in…
Comment linking /
on “GPT-4, without specialized training, beat a GPT-3.5 class model that cost $10M”Submitted Fine-Tuning Best Practices Series Introduction and Chapter 1: Training Data as a story · 3 points
HN discussionLots of folks working on open-source reasoning models trained with reinforcement learning right now. The best one atm appears to be Alibaba's 32B-parameter QwQ: https://qwenlm.github.io/blog/qwq-32b-preview/ I also recently wrote a blog explaining how reinforcement fine-tuning works, which is likely at least part of the pipeline used to train o1: https://openpipe.ai/blog/openai-rft
Comment linking /blog/openai-rft
on “A path to O1 open source”Submitted What we've learned in 3 days of Llama 3 as a story · 3 points
HN discussionThere are lots of third-party providers that will host your fine-tuned model for you, and just charge per token like OpenAI. Here are some of the providers I've personally used and would vouch for in production, along with their costs per 1M input tokens for Llama 3 8B, as a point of comparison: - Replicate: $0.05 input, $0.25 output - Octo: $0.15 - Anyscale: $0.15 - Together: $0.20 - Fireworks: $0.20 If you're look…
Comment linking /
on “Ask HN: How does deploying a fine-tuned model work”Submitted Fine-Tuning Best Practices: Models as a story · 2 points
HN discussionOk good questions here. By fine-tuning in this context I assume you mean "supervised fine-tuning", or SFT. SFT trains a model to produce a specific string of output tokens, given an input. With SFT, if you were trying to train an assistant to solve math problems using a code interpreter, you might train it on a dataset that looks like: input: 'What is 934+1208' output: `print(934+1208)` input: 'how many "r"s in stra…
Comment linking /blog/art-e-mail-agent?refresh=1746030513873
on “Show HN: ART – a new open-source RL framework for training agents”Submitted Fine-Tuning for Production Apps as a story · 2 points
HN discussionContributor here, we developed the Agent Reinforcement Trainer (ART) library to make it easy to train LLMs for anything. No callbacks or straitjacket flows. Instead we serve an OpenAI API-compatible endpoint that you can use as a drop-in replacement for any proprietary APIs you may be hitting. After collecting responses from the inference API, you can tune the model with your own custom rewards and repeat the proces…
Comment linking /blog/art-e-mail-agent
on “Show HN: ART – a new open-source RL framework for training agents”Submitted LLM Fine-Tuning Best Practices: Base Models Proprietary/Open Source, Large/Small as a story · 2 points
HN discussion
