The batch LLM platform

Run batch LLM workloads over millions of rows, on the right model, in your region.

Send any amount of data, or point Luemyn straight at your warehouse or lakehouse. Pick frontier, open or your own fine-tuned models, or let intelligent routing choose the right one for each prompt. Results go back to your tables, with zero data retention.

ClassificationExtractionEnrichmentLabellingSummarisationEvaluation
Join the early-access list Why Luemyn

Three problems, solved in one place

Batch LLM work should be the cheapest, simplest way to put models to work on your data. Today it isn't.

At any scale

Cheap or easy: today you pick one.

Batch APIs are the cheapest way to run frontier models, but the plumbing is yours: limits, retries, failed rows, and moving data in and out of your warehouse. SQL AI functions are easy, but you pay for it. On Snowflake, Claude Sonnet 4.6 costs 2.4× its batch API price, before warehouse compute.1

Luemyn: straight from your warehouse and back, built on batch-API pricing.

  • Send any volume. We handle splitting, rate limits, retries and failed rows.
  • Read from your warehouse or lakehouse, and write results back to your tables.
  • One API across frontier and open-model providers, with live progress and cost for every job.
On the right model

One model doesn't fit every row.

At billions of rows, the model you choose is the whole bill. Running a frontier model on every row overpays for the easy ones. Switching to a cheaper model is risky when you can't see whether quality holds, and fine-tuning your own is a project of its own.

Luemyn: frontier models, open models we host, and your own fine-tuned models in one pipeline.

  • Hosted open-weight models behind the same API.
  • Fine-tune a model for your task on your own data.
  • Intelligent routing sends each prompt to the right model, measured against your quality bar.
In your region

Your data can't leave.

Many teams can't send data to a service that keeps it, or that processes it outside their region.

Luemyn: we keep nothing, and EU data stays in the EU.

  • Zero data retention.2
  • EU region: on our hosted open and custom models, data never leaves the EU.

How a job runs

Bring a table and a definition of a good answer.

Your table→ Pick models, or let Luemyn route→ Run at batch scale→ Results back in your table

Rolling out with design partners

We're building in this order, and design partners get each step first.

First

Batch on frontier models, straight from your warehouse

Next

Hosted open models, with an EU region

Then

Fine-tuning your own models

Then

Intelligent routing across every model

Become a design partner

We're working with a small number of teams running large batch LLM workloads. Tell us what you're processing and where your data lives.

[email protected]

1. Snowflake AI_COMPLETE, claude-sonnet-4-6: 1.80 / 9.00 AI Credits per million input / output tokens at $2.00 per AI Credit ($3.60 / $18.00), Snowflake Service Consumption Table 6(a). Anthropic Message Batches API: $1.50 / $7.50 per million tokens, with results within 24 hours. Snowflake warehouse compute is billed separately. List prices, November 2026.

2. Luemyn stores no prompts or outputs once a job completes and results are delivered. When a job runs on a third-party frontier model, that provider's data terms also apply.