Cactus
Fine-tune Needle

Needle fine-tuning pricing

Local training is free. Hosted fine-tuning starts at $19.

On your hardware

Free

Fine-tune locally with the open-source package.

$0

No account or card required

  • Train on your own hardware
  • LoRA tuning with a frozen base
  • 4-bit model exports
  • Your own training data
  • Validation loss reporting
  • Open-source CLI workflow

Bring your own compute, datasets and storage. No platform allowances included.

Get started on GitHub
On Cactus GPUs

Starter

Try hosted fine-tuning.

$19once

30 days of access · No renewal

  • Hosted GPU compute included
  • Full-depth model training at every layer
  • 2-bit model exports
  • Confidence head trained on your tools
  • Your data augmented with Needle’s original dataset
  • Validation & test accuracy per depth

3 fine-tuning runs

3,000 generated examples

10 GB platform storage

Try Starter for $19
Long-term value
On Cactus GPUs

Pro

For ongoing hosted fine-tuning.

$99/ month

Billed monthly · Cancel anytime

  • Hosted GPU compute included
  • Full-depth model training at every layer
  • 2-bit model exports
  • Confidence head trained on your tools
  • Your data augmented with Needle’s original dataset
  • Validation & test accuracy per depth

10 fine-tuning runs / period

10,000 generated examples / period

10 GB platform storage

Choose Pro
On Cactus GPUs

Custom

Custom platform allowances.

Let’s talk

For higher-volume requirements

  • Hosted GPU compute included
  • Full-depth model training at every layer
  • 2-bit model exports
  • Confidence head trained on your tools
  • Your data augmented with Needle’s original dataset
  • Validation & test accuracy per depth

Custom run, data-generation and storage allowances. Contact us to scope your requirements.

Talk to us

Starter is available once per account before subscribing.