Product coverage
Traces
Datasets
Evals
Optimise
Train
Serve
Overmind
OpenPipe
sunset 30 Jul 2026
OpenPipe's coverage, shown as it stood at sunset. Since 30 July 2026 the platform accepts no new training or inference; RL continues inside W&B Training.
Which one fits you
Decision tree
Question 1 of 3, Where does the sunset leave you?
Where does the sunset leave you?
Overmind does not do reinforcement learning. Three of the answers here send you to ART and W&B Training instead of Overmind, and two more run them alongside.
Product comparison
| Overmind | OpenPipe | |
|---|---|---|
| Status | Live, in active development | Sunset 30 July 2026; site is an archive |
| Data capture | OpenTelemetry-native, any OTel exporter | Proprietary SDK wrapper or proxy. Never OTel |
| Observability | Spans, sessions, live scores, context graph from your repo | Request logs with metadata tags |
| Datasets | Two-pass audit, provenance, PII masking. JSONL imports | Filtered logs or JSONL upload, stronger-model relabelling |
| Training | LoRA or full fine-tuning, parallel experiments | SFT and DPO historically; RL via ART / W&B Training |
| Code optimisation | Scored diffs, opened as PRs | Not offered |
| Proof before switching | Benchmark vs your production model, overlap-checked | Head-to-head evals; production baseline was your job |
| Weights | Downloadable, always | Exportable for open models; successor serves LoRA via W&B |
Common questions
Is OpenPipe still available?
Not as a standalone platform. CoreWeave acquired it in September 2025, and openpipe.ai stopped accepting new training and inference on 30 July 2026. The site describes itself as an archive; the technology continues inside W&B Training. ART, the open-source RL trainer, remains active.
Can I import my OpenPipe datasets into Overmind?
Yes. They are OpenAI-format JSONL, which the Data Workshop imports directly. Imported rows get the same treatment as native ones: quality audit, provenance, versioned commits, PII masking.
Does Overmind do RL like ART?
No. Overmind trains with LoRA and full fine-tuning, benchmarked against your production baseline. For GRPO-style RL over multi-step trajectories, ART and W&B Training are the right tools, and they can sit alongside Overmind's observability and datasets.
How does pricing compare?
OpenPipe's per-token rates stopped applying when the platform closed to new work in July 2026; its successor prices through Weights & Biases. Overmind: Free and Pro subscriptions plus credits, with trace ingest never credit-gated. As of August 2026.