Cerebrium
Free tierServerless GPU infrastructure for real-time AI with sub-second cold starts and instant autoscaling
Key strengths
Facts verified against cerebrium.ai on
Recent changes
- Free tier: now available
- Plans updated: Pay-per-second compute pricing. Hobby plan is free (+ compute costs). Standard plan is $100/month + compute. Enterprise is custom. GPU rates range from $0.000164/s (T4) to $0.00167/s (B200). Memory at $0.00000222/GB/s, storage at $0.05/GB/mo (first 100GB free).
Get alerts when Cerebrium changes
Free. We'll email you when pricing, the free tier, API access or open-source status changes.
Cerebrium — Serverless GPU Infrastructure for Real-Time AI
Cerebrium is a serverless GPU compute platform purpose-built for deploying production AI workloads — LLMs, voice agents, and image/video models — without managing infrastructure. Launched in 2023, it offers pay-per-second billing with no idle costs, elastic autoscaling across 12+ GPU types (T4 through B200), and sub-second to low-second cold starts powered by GPU and memory snapshotting.
Key technical characteristics:
- Deployment: Bring your own code or Dockerfile — no infrastructure rewrites required
- Scaling: Elastic autoscaling across multiple regions with 99.999% uptime and multi-region failover
- Observability: OpenTelemetry-native instrumentation built in
- Compliance: SOC 2, HIPAA, GDPR, and ISO 27001 certified
- API: Full API available for programmatic control
- Open source: Yes (528 GitHub stars)
Integrations: Twilio, Pipecat, LiveKit, vLLM, SGLang, Triton Inference Server, TensorRT-LLM, WandB, Gradio, FastAPI, OpenTelemetry
Pricing: Usage-based. Hobby plan is free (+ compute costs). Standard plan starts at $100/month + compute. Enterprise is custom. GPU rates range from $0.000164/s (T4) to $0.00167/s (B200); memory at $0.00000222/GB/s; storage at $0.05/GB/mo (first 100 GB free).