Baseten
Production inference for custom and open-source models — dedicated GPUs, serverless scale.
About
Baseten serves custom and fine-tuned models on dedicated single-tenant GPUs, packaged through its open-source Truss framework — weights, model code, and hardware config in one deployable unit. Alongside dedicated deployments it offers per-token Model APIs for popular open-weight models, a Chains SDK for orchestrating multi-model workflows (voice AI, agents, RAG), specialized engines like TensorRT-LLM, and cold-start optimizations. Security depth is serious for its size: SOC 2 Type II, HIPAA, GDPR, PCI DSS. In December 2025 it signed a strategic collaboration agreement with AWS.
The honest picture: dedicated replicas bill continuously until you scale them down, and per-hour GPU pricing runs premium versus commodity clouds. The model catalog is open-weight only — teams that also want GPT or Claude need a second vendor.
What it can do
- Truss open-source model packaging
- Dedicated single-tenant GPU deployments
- Chains multi-model orchestration SDK
- Per-token Model APIs for open models
- Autoscaling with cold-start optimization
Setup
Sign up at baseten.co. Deploy with truss push, or query Model APIs instantly with an API key.
Prerequisites
Model weights for dedicated deploys (or just use Model APIs). A Baseten API key.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 51; Trending score 76/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.