Free GPU quotas
How much free GPU time each service gives you, how long a session can run and when the quota resets, with the vendor page behind every number. Pair it with the free GPU comparison and the free GPU guide, or let the stack builder pair free GPU hours with a host and an LLM API.
Checked Oct 13, 2026 · re-checked monthly
Hours left this week
Your dashboard shows this; it stays in your browser.
30 of 30 hours left
Weekly, Saturday 00:00 UTC (Kaggle's floating-quota announcement) Floating quota: Kaggle says 30 hours is sometimes higher depending on demand.
Published free GPU allowances
| Service | Free allowance | Max session | Resets | GPU / TPU |
|---|---|---|---|---|
| Kaggle notebooks | 30 GPU hours a week | Not published | Weekly, Saturday 00:00 UTC (Kaggle's floating-quota announcement) | TPU: 20 hours a week, up to 9 hours per session |
| Google Colab (free tier) | Not published | 12 h | Not published | TPU: Available, not quantified: the GPU and TPU types vary over time |
| Hugging Face ZeroGPU (Spaces) | 5 minutes a day | Not published | Rolling: resets 24 hours after your first GPU use of the day | Half NVIDIA RTX Pro 6000 Blackwell (48 GB), Full NVIDIA RTX Pro 6000 Blackwell (96 GB, 2x quota cost) |
| Lightning AI (free plan) | Not published | Not published | Not published | Not named |
| Paperspace Gradient (free plan) | Not published | 12 h | Not published | Not named |
Notes and sources
- Kaggle notebooks: Floating quota: Kaggle says 30 hours is sometimes higher depending on demand. Phone verification unlocks accelerators. Docs pages render client-side; figures read via Kaggle's own pages in search, re-check in a browser at the monthly pass. Source (“weekly GPU quota ... 30 hours or sometimes higher depending on demand and resources”)
- Google Colab (free tier): Usage limits are dynamic and unpublished; resources are not guaranteed and idle runtimes time out (duration not stated). Source (“Colab does not publish these limits, in part because they can vary over time.”)
- Hugging Face ZeroGPU (Spaces): Free accounts: 5 minutes a day (2 minutes signed out, 40 on PRO). Default call duration 60 s; Gradio Spaces only; hosting needs a verified email and an account older than 30 days. Source (“Included daily quota resets exactly 24 hours after your first GPU usage.”)
- Lightning AI (free plan): Vendor pages did not render on 2026-10-13 and lightning.ai search snippets conflict (30 starting credits / about 80 GPU hours vs 15 monthly credits), so no number is recorded here; see the offer page for the verified terms. Source (“up to 30 credits (~80 GPU hours) to start”)
- Paperspace Gradient (free plan): Free plan covers basic instances only (GPU model not named), public projects only, 5 GB storage, no availability guarantee. Source (“Auto-shutdown (12 hour limit)”)
Understand it
A few minutes of play-along, no sign-up.
- How did it learn all this? · Training: loss and gradient descent · 8 min
- How does a 30 GB model fit on my laptop? · Quantization (bits per weight) · 5 min