Honest pricing

Per-second billing. No surprises.

We publish every price — including where we're more expensive than hyperscalers. Updated quarterly.

Serverless inference

Per million tokens. Per image. Per minute. No minimums.

Model classWattlendHyperscaler comparableYou save
Llama-3-8B (or equiv)$0.20 / $0.40 per M tok$0.30 / $0.60~35%
Llama-3-70B (or equiv)$0.55 / $1.20 per M tok$0.99 / $2.50~50%
Mixtral 8×7B$0.25 / $0.50 per M tok$0.40 / $0.80~38%
Whisper-large-v3$0.0030 / min$0.0060 / min~50%
SDXL 1024×1024$0.0035 / image$0.0080 / image~56%
FLUX.1 [dev]$0.0080 / imagen/a
Custom container$0.30 – $0.85 / GPU-hr$0.80 – $4.00 / hr~50%

On-demand dedicated GPU

Reserve the whole GPU. Pay only for time used. Per-second metering.

GPUVRAM$ / hour
RTX 409024 GB$0.45
RTX 509032 GB$0.65
RTX A600048 GB$0.70
RTX 6000 Ada48 GB$1.10
H100 80GB80 GB$2.50

Spot (preemptible) & Reserved

Spot
~50% of on-demand

Same GPU, can be preempted with 60-second notice. Great for batch jobs, dev workloads, anything restartable.

Reserved
up to 35% off

1-month commit: 15% off. 12-month commit: 35% off. Grandfathered at signup rate.

Enterprise SLA

$5,000+/mo retainer + usage
99.9% SLA · SSO · audit logs · private pool · custom contract

Includes dedicated support and named account manager. Compliance: SOC 2 Type II (Phase 2), ISO 27001 (Phase 3), HIPAA BAA on request.

What we don't charge for

  • Egress bandwidth — free up to 100 GB per endpoint per month (then $0.02/GB).
  • API requests — free.
  • Logs and metrics retention — 30 days free.
  • Pay-as-you-go from your first request.
  • $200 more credit when you ship your first real workload within 7 days.
  • No idle endpoint fees on the serverless tier.

Where we're more expensive than hyperscalers

We say this on our pricing page on purpose. Trust beats spin.

  • !p99 cold-start latency on serverless is currently worse than AWS Bedrock. Improving quarterly.
  • !Latest GPUs (B200, MI300X) at scale are easier to get on hyperscalers right now.
  • !Some enterprise integrations (AWS PrivateLink, VPC peering) aren't supported yet — Phase 3.

Operator revenue split

GPU contributors take home 70–80% of what their hardware earns. The rest covers payments, support, control-plane infra, fraud loss, and R&D.

TierQualificationOperator share
Cold-startfirst 30 days as operator80%
Bronzedefault70%
Silver30-day uptime ≥ 95%72%
Gold90-day uptime ≥ 99%75%
Platinum180-day uptime ≥ 99.5%, top 5%78%
© 2026 Wattlend. All rights reserved.