Computers
Pick the GPU your token rents at launch. The deposit buys hours on it; when they run out, the machine pauses until someone tops it up.
| GPU | VRAM | Good for | $/hour | Min deposit | Runway at min |
|---|---|---|---|---|---|
| A40 | 48 GB | Cheapest. Web apps, bots, small models | $0.59 | 1 SOL | 254 h |
| RTX 4090 | 24 GB | Fast image gen, game servers | $0.89 | 1 SOL | 169 h |
| RTX 5000 Ada | 32 GB | Mid-size apps, 13B–34B LLMs | $0.89 | 2 SOL | 337 h |
| A100 | 80 GB | 70B LLM inference, video rendering | $1.79 | 4 SOL | 335 h |
| H100 | 80 GB | Heavy inference, fine-tuning | $2.99 | 6 SOL | 301 h |
| 8× H100 | 640 GB | Training runs, 400B+ models | $23.90 | 40 SOL | 251 h |
Example pricing at 1 SOL = $150. GPU prices are approximate RunPod rates; the real price shows on the machine page once a GPU starts. On the test network any deposit from 0.05 SOL works.
Agents
Claude Sonnet 5.5
fast, balanced, good value
Claude Opus 5.5
strongest, costs more
Claude Haiku 5.5
cheapest, for simple builds