OpenPipe
RL fine-tuning platform, now folded into CoreWeave
Overview
OpenPipe turned production LLM logs into cheaper fine-tuned models and built ART, an open-source agent RL trainer. Acquired by CoreWeave in September 2025; its legacy self-serve platform stops taking new training/inference on 30 July 2026, migrating to Weights & Biases.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with OpenPipe
- Captured production LLM request logs as fine-tuning data
- Fine-tunes smaller open-weight models to replace GPT-4-class calls
- ART: open-source GRPO trainer for multi-step agents
- Managed SFT (supervised fine-tuning) workflows
- Hosted per-token and hourly-compute inference
- Now backed by CoreWeave's GPU infrastructure
ASEAN Perspective
OpenPipe in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
OpenPipe made its name capturing production LLM traffic and distilling it into cheaper fine-tuned models, then pushed into agent reinforcement learning with ART (Agent Reinforcement Trainer), its open-source GRPO training library for teaching multi-step agents from their own trajectories. That combination — practical distillation plus a genuinely useful open-source RL tool — earned it real developer mindshare and a September 2025 acquisition by CoreWeave.
The catch for anyone evaluating this listing today: the original self-serve OpenPipe platform is being retired. New training and inference on the legacy product stop 30 July 2026, with workloads migrating to Weights & Biases (also now a CoreWeave company). ART and managed SFT keep going under the CoreWeave umbrella, so the technology survives, but 'sign up at openpipe.ai and fine-tune a model' is no longer really the product. Treat this as an infrastructure component of CoreWeave now, not a standalone SaaS to evaluate on its own.
What people say
There's no meaningful independent review trail for OpenPipe on G2, Capterra or the usual SaaS review sites — searches turn up SEO listicles restating the marketing copy, not verified customer feedback. That's not unusual for a developer infrastructure tool sold mostly through direct outreach and Y Combinator/Twitter word of mouth, but it means this write-up leans on what's publicly documented rather than a chorus of user praise or complaints.
What is documented: OpenPipe raised a modest $6.7M seed in March 2024 (Costanoa Ventures, Y Combinator, with angels including GitHub co-founder Tom Preston-Werner and Copilot co-creator Alex Graveley) before CoreWeave announced its acquisition in September 2025, folding OpenPipe in alongside its earlier purchase of Weights & Biases to build out a training and reinforcement-learning stack. CoreWeave's press materials call OpenPipe's ART library 'the most widely used open-source library for training agents with RL' — a claim from the acquirer, worth some skepticism, though the GitHub repo shows active releases and real community use, and OpenPipe's own benchmark (ART-E) claims a fine-tuned Qwen 2.5 14B email agent beating OpenAI's o3 on that specific task.
Pricing for the still-running legacy platform is straightforward and usage-based: training runs $0.48 per million tokens for 8B-class models up to $2.90/million for 70B+, hosted inference bills per-token for high-volume models or by the hour ($1.50-$12/hr) for lower-volume compute units, with enterprise volume discounts available on request. But that pricing is now on a clock — the legacy self-serve product stops accepting new training and inference work on 30 July 2026, with existing workloads pointed at Weights & Biases. Anyone evaluating OpenPipe today should really be evaluating ART (still open-source, still maintained) and CoreWeave's managed SFT offering, not the original SaaS product. Existing customers cited in OpenPipe's own case studies include Notion and Sourcegraph, both high-volume LLM shops with the kind of repetitive-query workloads distillation suits best — a reasonable proof point, though drawn from the company's own marketing rather than an independent case study.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including OpenPipe's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with OpenPipe unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to OpenPipe directly →
Spotted something out of date? Suggest an update →
More in Code & Dev Tools
