Every account gets 10 small fine-tuning runs each month, no card required. Upgrade to Pro for recurring compute credits and use prepaid credits for larger jobs.
Failed and cancelled paid runs are free. Specs, evals, downloads, and local serving are free during beta.
$0.00
For community projects, students, open-source examples, and early experiments.
$9.00/mo
For builders running more experiments, auto-tune loops, and private workloads.
Usage
For large fine-tunes, teacher labeling, and Pro auto-tune usage without surprise bills.
Price per 1M training tokens, per epoch.
google/gemma-4-E2B-it
Size
E2B
Price
$0.45
Qwen/Qwen3.5-2B
Size
2B
Price
$0.45
ibm-granite/granite-3.3-2b-instruct
Size
2B
Price
$0.45
Qwen/Qwen3-VL-2B-Instruct
Size
2B VL
Price
$0.55
meta-llama/Llama-3.2-3B-Instruct
Size
3B
Price
$0.55
bigcode/starcoder2-3b
Size
3B
Price
$0.55
google/gemma-4-E4B-it
Size
E4B
Price
$0.70
Qwen/Qwen3.5-4B
Size
4B
Price
$0.70
microsoft/Phi-4-mini-instruct
Size
3.8B
Price
$0.70
Rates may change as provider pricing changes. Check the docs for billing details.
Free quota covers small monthly runs. Larger jobs use included Pro credits or prepaid credits, with estimates before launch.
Usually under $2
Validate the spec and dataset shape.
$10-$30
Fine-tune and evaluate a real task.
Estimated upfront
Iterate until the score target or cap is reached.
No card required for Free. Upgrade when you need recurring compute credits or use top-ups when you need more training credit.