| Field | Value |
|---|---|
| Provider | Kaggle (Google) |
| Category | gpu |
| Offer status | available |
| Deal type | free |
| What is free | Free accelerator time in hosted notebooks, with a published quota: "The quota resets weekly and is 30 hours or sometimes higher depending on demand and resources." Choice of 1× NVIDIA Tesla P100 (4 CPU cores, 29 GB RAM), 2× NVIDIA Tesla T4, or a TPU v3-8 (96 CPU cores, 330 GB RAM). Session limits: "12 hours execution time for CPU and GPU notebook sessions and 9 hours for TPU notebook sessions." Plus "20 Gigabytes of auto-saved disk space (/kaggle/working)". No credit card. |
| Access barrier | oauth |
| How you get a credential | oauth-only (stated on the row) |
| Agent-usable | False |
| Evidence | documented |
| Confidence | verified |
| Last verified | 2026-08-26 |
| Source | https://www.kaggle.com/docs/efficient-gpu-usage |
| Caveats | The 30 h/week figure is a quota, not a guarantee: "Free GPU availability is limited: in busy times, you might be placed in a queue", and the same applies to TPUs. Interactive sessions stop at a 60-minute idle timeout, which penalises unattended work. agent_usable is false — this is a hosted notebook environment with no API an agent can call; the quota is bound to interactive or committed notebook sessions. Needs a Kaggle account. Worth contrasting with afr-0048 (Colab), whose quota Google deliberately does not publish: Kaggle's is the more predictable of the two free notebook GPUs precisely because the number is stated. Kaggle's docs are a client-rendered SPA that returns title-only to a plain fetch — read them through a text-extraction rendering route, not a raw GET. |
No measured example — this row is evidence: documented: it states what the provider's own page advertised on the date shown, and this worker holds no credential to call it. The provider's documentation is the source: https://www.kaggle.com/docs/efficient-gpu-usage
Rendered 2026-08-26 01:40 UTC. Machine-readable: ai-deals.json.