Acces GPU gratuit illimite — Cloud + Local + LPU/WSE — 0 EUR total
bash /opt/runpod-launch.sh
T4 16GB / L4 24GB / A100 40GB — ~12h/jour gratuit. Notebooks Python, PyTorch, TF
T4 x2 / P100 — 30h/semaine gratuit. Competitions, datasets, notebooks
L4 / A10G — 22h/mois gratuit. Studios ML, deploy models
T4 16GB ZeroGPU — Inference endpoints gratuits, Qwen2.5-72B router
A10 24GB — Always Free tier, VM GPU persistante
llama-3.3-70B — Custom LPU chips, <1s, primary cascade
Qwen-3-235B — Wafer-Scale Engine, ultra-rapide
Nemotron-49B — UNLIMITED, zero rate-limit, 2-5s
DeepSeek-V3.1 — RDU dataflow, free tier (balance 0)
Llama 3.1 8B + DeepSeek R1 32B — GPU edge, 10K neurons/jour
Google 2026 #1 open-source — 9.6GB, function calling, multimodal. Apache 2.0
Alibaba — 2.5GB, rapide, multilingue FR/AR/EN
Embeddings RAG — 274MB, Qdrant ingest
Embeddings light — 45MB, fast similarity