VMs + inference

# The cost of   a complete fleet.

Explore the inference and environment costs behind a run, alongside the Multiplet monthly plan.

## The whole run,  handled together.

Multiplet brings model inference, the harness, working computers and fleet operation into one service. This calculator separates the rented components from that complete offer, so you can explore the economics for a workload.
[First fleet: 256 agents × four hours · $1,000](/first-fleet/) 

Cost calculator 

## Set the workload.

Adjust the agents and time. A standard agent-hour is roughly one credit.

Agents

1 10,000 
642565121,000

Duration

hours

days

1 hour  48 hours 

Estimated run usage 
8,000credits 
Usage value at the monthly plan rate $3,906.25
Drawn from the monthly allowance. This is not a standalone run price.

1,000 agents × 8 hours

About one credit per agent-hour in the standard configuration. Model choice, larger computers and delivered service affect final usage. Your agent checks the estimate before launch.

The component comparison 

## VMs + inference.

What would the parts cost if you rented them separately? Use the fleet size and duration above, then change the assumptions below.

GLM-5.3 inference

Prime Intellect via OpenRouter

DeepInfra via OpenRouter (promotional)

Baseten FP4 via OpenRouter

Rented Daytona VM · per agent

Sizes for the rented-components estimate. Multiplet environments are configured separately.

Daytona · 1 vCPU · 2 GiB RAM · 16 GiB disk

Daytona · 2 vCPUs · 4 GiB RAM · 16 GiB disk

Input served from cache
93.6%
No cache hits All input cached Use the recorded cache rate 

1,000 agents × 8 hours

GLM-5.3 inference
$1,272

Daytona environments
$672

Rented components
$1,944

Monthly plan · usage value
$3,906

The Multiplet figure values the estimated credits used from the monthly allowance, not a standalone purchase. The plan is $10,000/month. Rented components cover model calls and computers; building and operating the fleet takes additional work.

Workload, rates and assumptions

Token usage scales from our recorded 2,048-agent run by configured agent-hours. The measurement includes degraded service and ended in a crash. This estimates a token workload; it does not establish equivalent throughput, task results or a provider’s capacity.

VMs run for the full duration. Storage above Daytona’s first 5 GiB is charged. Router fees, network, plan fees, discounts, taxes and operating work are excluded. No paid cache-retention fees are modeled. Cache reuse is an assumption you can change, not a measured remote result.

Rates checked 16 October 2026. DeepInfra’s listed rate is promotional. Endpoint configurations and quantization can differ.
[Download the calculation inputs](/evidence/remote-cost-model.json) [Workload data and conditions](/evidence/2048-agent-run.json) [Provider token prices](https://openrouter.ai/z-ai/glm-5.3) [Daytona prices](https://www.daytona.io/pricing) 

Provider coverage and additional component prices

The calculator uses recorded token demand, GLM-5.3 endpoints and Daytona environments. Rates are an October 16 snapshot. Component estimates exclude the engineering and operating work that Multiplet supplies.

Boat provides another environment option: its listed default 4-vCPU, 8-GB VM is $0.036/hour, or $36.86 for 256 environments used for four hours. The $100 tier includes compute credit and 300 concurrent environments; model usage is separate. It is a different resource configuration. These are component estimates rather than matched service-performance comparisons.
[Boat pricing · October 18](https://docs.boat.dev/pricing) [What Multiplet operates](/compare/build-your-own/)