What agent tasks cost
Tokens priced at Opus 5.5 rates ($5 per 25-minute session). Tokens are most of the bill, so model choice matters more than machine size.
Monthly token cost by model
10-minute tasks, billed over 30 days. Same token usage for every model, priced at each model's rates.
| Tasks/day | Opus 5.5 | Kimi K3 | Sonnet 5.5 | DeepSeek-V3.2 |
|---|---|---|---|---|
| 25 | $1,500/mo | $1,215/mo | $810/mo | $178/mo |
| 50 | $3,000/mo | $2,430/mo | $1,620/mo | $356/mo |
| 75 | $4,500/mo | $3,645/mo | $2,430/mo | $533/mo |
| 100 | $6,000/mo | $4,860/mo | $3,240/mo | $711/mo |
| 150 | $9,000/mo | $7,290/mo | $4,860/mo | $1,067/mo |
| 200 | $12,000/mo | $9,720/mo | $6,480/mo | $1,422/mo |
| 250 | $15,000/mo | $12,150/mo | $8,100/mo | $1,778/mo |
| 300 | $18,000/mo | $14,580/mo | $9,720/mo | $2,133/mo |
- Rates per minute of session time: Opus 5.5 $0.20, Kimi K3 $0.162, Sonnet 5.5 $0.108, DeepSeek-V3.2 $0.024.
- Token costs only. For compute, add $63/mo per 25 tasks/day on 8 GB machines, or $126/mo on 16 GB.
Blended cost calculator
Set your monthly volume, average task length, model mix and machine size. Opus 5.5 takes whatever share the other models don't.
900 Opus, 900 Kimi, 1,200 Sonnet and 0 DeepSeek tasks per month.
$1.60 per task, about 100 tasks per day
| Opus 5.5 tokens | $1,800/mo |
| Kimi K3 tokens | $1,458/mo |
| Sonnet 5.5 tokens | $1,296/mo |
| DeepSeek-V3.2 tokens | $0/mo |
| Compute (8 GB) | $252/mo |
| Total per month | $4,806/mo |
Blended token rate $0.152/min. All-Opus would be $6,252/mo, so this mix saves $1,446/mo.
Not sure which model fits which task?
We'll help you pick the right model for each job, route the rest to cheaper ones, and size the compute, so you pay for quality only where it counts.
Compute: AWS Lambda MicroVMs, us-east-1, Graviton. 8 GB / 4 vCPU at about $0.504/hr and 16 GB / 8 vCPU at about $1.009/hr. Excludes bursting, snapshots and data transfer.
Tokens: a 25-minute high-complexity session of about 2.0M cached input, 0.22M uncached input and 0.19M output tokens, split evenly by time. Opus 5.5 at $4 / $20 per 1M ($0.20 cache read), Kimi K3 on AWS Bedrock (Global Standard) at $3 / $15 ($0.30 cache read), Sonnet 5.5 at $2 / $10 ($0.20 cache read), DeepSeek-V3.2 on AWS Bedrock (US regions) at $0.62 / $1.85, with the $0.056 Google Cloud cache-hit rate used where Bedrock lists none. Estimates only; actual usage varies by task.