Instant deploys
Push a model from Hugging Face, S3, or a local checkpoint and get a production endpoint in under two minutes. No Dockerfiles, no CUDA archaeology.
Series B · Backed by Meridian Ventures
Run any model.
Run any model.
At light speed.
Lumen Labs is the inference layer for production AI. One API to deploy, scale, and observe any open or fine-tuned model across a global GPU mesh — with cold starts under 900 ms and p99 latency you can put in an SLA.
4.2B+tokens served daily
<900mscold start, any model
99.99%uptime, 12-month trailing
31edge regions worldwide
The platform
Everything between your model and your users
Stop stitching together queues, autoscalers, and dashboards. Lumen ships the whole inference stack — hardened, observable, and boringly reliable.
Autoscale to zero
Scale from zero to thousands of concurrent requests and back. You pay for compute measured in milliseconds, not idle GPUs measured in regret.
Global edge routing
Requests land on the nearest of 31 regions and ride our private backbone to warm capacity. Your Sydney users stop paying the Virginia tax.
Deep observability
Per-request traces, token-level latency breakdowns, drift alerts, and cost attribution by team, feature, and prompt — built in, not bolted on.
Enterprise-grade security
SOC 2 Type II, HIPAA-ready, private networking, and single-tenant GPU pools. Your weights never leave your boundary — we route, you own.
Zero lock-in
OpenAI-compatible endpoints, Terraform provider, and one-command export of every artifact. Leave whenever you like — teams rarely do.
How it works
From checkpoint to production in three moves
The average team ships their first Lumen endpoint before their coffee goes cold. Here is the whole ceremony.
01 / DEPLOY
Point us at your model
One CLI command or API call. We handle quantization, batching strategy, and GPU placement automatically.
$ lumen deploy llama-3-70b \
--gpu h100 --regions auto
✓ endpoint live in 94s
02 / INTEGRATE
Swap one base URL
Endpoints speak the OpenAI wire format. Your existing SDKs, agents, and eval harnesses work unchanged.
client = OpenAI(
base_url="lumen.internal/v1",
api_key=env.LUMEN_KEY
)
03 / SCALE
Ship and watch it hold
Traffic spikes route to warm replicas across the mesh. Dashboards, alerts, and cost caps are on by default.
p50 212ms p99 748ms
rps 14,208 ▲ autoscaled x38
cost $0.11 / 1M tokens
Customers
Trusted by teams who cannot afford a slow token
From seed-stage agents to Fortune 500 copilots, over 1,400 engineering teams run their inference on Lumen.
We moved 100% of production inference to Lumen in a weekend. p99 dropped 61% and our GPU bill dropped 43%. I kept re-checking the dashboard because I didn't believe it.
Maya OkonkwoVP Engineering, Vantage AI
Black Friday hit our copilot with 40x normal traffic. Lumen autoscaled across three regions and nobody on my team got paged. That's the entire review.
Daniel ReyesHead of Platform, Nexbyte
The token-level tracing paid for itself in week one — we found a prompt regression that was silently costing us $11k a month. No other provider even shows you this.
Sarah ChenCTO, Driftwave
Pricing
Pay for tokens, not for idle silicon
Every plan includes the full platform — routing, autoscaling, and observability. You only choose how much throughput and control you need.
Build
For prototypes and side projects that might become companies.
$0/ month
- 5M free tokens every month
- All open models, serverless GPUs
- Community support
- 3 concurrent endpoints
Scale
For products in production with real users and real latency budgets.
$499/ month + usage
- From $0.11 per 1M tokens
- Fine-tuned + custom model hosting
- 99.9% SLA with p99 latency targets
- Unlimited endpoints, 31 regions
- Slack support, 4-hour response
Enterprise
For regulated workloads, single tenancy, and negotiated throughput.
Custom
- Dedicated GPU pools & VPC peering
- SOC 2, HIPAA, custom DPAs
- 99.99% SLA, 24/7 on-call engineers
- Committed-use pricing from $0.06/1M
Get started
Your first endpoint is 94 seconds away
Sign up, deploy a model, and send your first request on 5M free tokens. No credit card, no sales call, no YAML.
Free tier forever · SOC 2 Type II · Cancel anytime