DeepInfra — Inference cloud serving 200+ open-source (and some proprietary) AI models via API, plus dedicated GPU deployments.. Pricing model: Pure usage-based pay-as-you-go (prepaid or invoiced), no commitments; service-tier multipliers. Value metric: Tokens (LLMs), execution time, images, audio minutes; GPU-hours for dedicated.
At a glance
| Company | DeepInfra |
|---|---|
| What it does | Inference cloud serving 200+ open-source (and some proprietary) AI models via API, plus dedicated GPU deployments. |
| Headquarters | USA (city not verified) |
| Pricing model | Pure usage-based pay-as-you-go (prepaid or invoiced), no commitments; service-tier multipliers |
| Value metric | Tokens (LLMs), execution time, images, audio minutes; GPU-hours for dedicated |
| Pricing page | deepinfra.com/pricing |
Plans and prices
Per 1M tokens e.g. Llama-3.1-8B $0.02/$0.04, DeepSeek-V3.2 $0.26/$0.38; GPUs A100 $1.49/hr to B300 $5.89/hr; Standard 1x, Priority 1.5x, Flex 0.8x. No free tier, no minimum.
Growth milestones
DeepInfra appears in our analysis of 2026's $100M ARR class. Figures are company-reported unless stated.
- 2026-05: $107M Series B (valuation not disclosed). (source)
- 2026-09-29: Surpassed $100M ARR ($105M run rate), 14x YoY, 22 trillion tokens per week; customer LiveKit. (source)
How it grows
Developer self-serve with lowest-price positioning on open models; fast availability of new open models; usage scales with customers' AI products (LiveKit).
What you can steal
usage-based pricing
Offer speed tiers as price multipliers (priority 1.5x, flex 0.8x): customers choose the trade-off and you monetise urgency.
pricing model
Prepaid balances with spending limits reduce bad debt and reassure customers about runaway bills.
Sources

Get both books free
The Subscription Playbook (ebook + audiobook) and Feed a Starving Crowd, instantly. In Australia? Claim the free hardcover.
