
80 m
What does my agent cost?
An agent pays for its brain by the hour and for its hands by the transaction. Set yours up and Shello prices both at this minute's live rates.

Brain and hands, per day
precisionHow finely the model's numbers are stored. int4 fits a smaller, cheaper GPU. More in the glossary
urgency
Your agent costs about $3.68 a day, $110.50 a month.
Most of it is the GPU it thinks on. GPU: T4 on Akash, $0.15 an hour, .
| part | per day | per month |
|---|---|---|
| GPU | $3.60 | $108.00 |
| priority fees | $0.08 | $2.50 |
| Jito tips | $0.00 | $0.00 |
Holders can save this setup; Shello recomputes it every day.
GPU or API?
At this volume an API is cheaper: $0.66 a day (Groq) against $3.60 for a GPU running all day.
Show the API comparison (for developers)
- $ / 1M tokens
- list price per million tokens, blended 75% input and 25% output
- breaks even at
- the daily tokens at which a GPU running 24 hours costs the same as the API
| option | $ / 1M tokens | per day | GPU breaks even at |
|---|---|---|---|
| GPU: T4 on Akash, 24 h (read 06:02:01 UTC) | - | $3.60 | - |
| Groq openai/gpt-oss-20b (list price 2026-11-09) | $0.13 | $0.66 | 27.4M tokens/day |
| Anthropic Claude Haiku 5.5 (prompts up to 100,000 tokens) (list price 2026-11-09) | $0.20 | $1.00 | 18M tokens/day |
| OpenAI gpt-6-luna (list price 2026-11-09) | $0.20 | $1.00 | 18M tokens/day |
| Together AI gpt-oss-120b (openai/gpt-oss-120b) (list price 2026-11-09) | $0.26 | $1.31 | 13.7M tokens/day |
| Fireworks OpenAI GPT OSS 120B (list price 2026-11-09) | $0.26 | $1.31 | 13.7M tokens/day |
| Groq openai/gpt-oss-120b (list price 2026-11-09) | $0.26 | $1.31 | 13.7M tokens/day |
| Together AI Qwen2.5 7B Instruct Turbo (list price 2026-11-09) | $0.30 | $1.50 | 12M tokens/day |
| Together AI DeepSeek V4.1 Flash (list price 2026-11-09) | $0.53 | $2.62 | 6.9M tokens/day |
| Fireworks DeepSeek V4.1 Flash (list price 2026-11-09) | $0.53 | $2.62 | 6.9M tokens/day |
| Together AI Qwen3.7 Plus (list price 2026-11-09) | $0.56 | $2.80 | 6.4M tokens/day |
API price blended at 75% input and 25% output tokens. Whether one shell can serve this many tokens a day depends on its throughput, which Shello does not measure.