Hosted notebooks and compute with a free GPU or notebook tier, taken from each vendor's official page. Free GPU access is usually limited, queued or credit-based; see the catch for each.
5 listings · facts checked on official pages, latest 2026-10-20. Plans change; confirm on the vendor's site.
Hosted Jupyter notebooks free of charge, with very limited access to GPUs and standard-memory VMs.
How to get it
Open colab.research.google.com and sign in with a Google account.
What it requires
A Google account.
The catch
Free GPU access is heavily restricted and not guaranteed, notebooks run for at most 12 hours, and execution stops when you stop interacting; faster GPUs need compute units (Pay As You Go, Colab Pro or Pro+).
Up to 30 credits, 1 free active Studio (4-hour restarts), 32-core CPU Studios, single GPUs (L40S, A100, H100, H200) up to 2 at a time, and 200 GB of persistent storage.
How to get it
Choose Free on lightning.ai/pricing.
The catch
GPU time is pay as you go once the credits are used, and the free Studio restarts every 4 hours.
$30 in free compute credits every month on the Starter plan, 3 workspace seats, 100 containers with 10 GPU concurrency, scheduled and web functions (limited), and 1 TiB of free network egress a month.
How to get it
Get started at modal.com.
The catch
Usage beyond the $30 monthly credit is billed as compute, and egress past 1 TiB is $0.04 per GiB (from October 1, 2026); Team ($250/mo) is needed for more seats, containers and credits.
Hub hosting for models, datasets and Spaces, with Spaces on free CPU Basic (2 vCPU, 16 GB) and ZeroGPU hardware: 5 minutes of ZeroGPU time a day on a free account, and up to 2 hosted ZeroGPU Spaces.
How to get it
Sign up at huggingface.co.
What it requires
A free Hugging Face account.
The catch
Free accounts get no included Inference Providers credits; you buy credits or upgrade to PRO ($9/month with $2 of credits a month).