Overview
Baseten is an inference platform with two halves. Model APIs are per-token endpoints for popular open-weight models, such as GLM-5.3, DeepSeek V4.1 Flash and Kimi K3, that Baseten has already optimised. Dedicated deployments let you run your own, fine-tuned or open models on reserved GPUs billed per minute.
The Basic plan has no monthly fee: you pay per token or per GPU-minute (an H100 is $0.10833 a minute). Pro adds priority access to scarce GPUs and volume discounts, and Enterprise adds self-hosted or hybrid deployment in your own VPCs and custom SLAs. New accounts get some free credit to try deployments.
Baseten also offers training through its Loops SDK, so a model can be trained and then deployed on the same stack. It states SOC 2 Type II certification and HIPAA compliance, which matters for healthcare and regulated customers. For a small project, the per-token Model APIs are the practical entry point; dedicated GPUs only pay off once traffic is steady enough to keep an instance busy.
Pricing and plans
| Plan | Price | What you get |
|---|---|---|
| Basic | $0/mo | Pay as you go |
| Pro | Custom | Priority access to high-demand GPUs and volume discounts |
| Enterprise | Custom | Self-hosted or hybrid, custom SLAs |
| GLM-5.3-Flash | $0.15 in / $0.50 out per 1M | Cached input $0.03 |
| DeepSeek V4.1 Flash | $0.30 in / $1.20 out per 1M | Cached input $0.007 |
| Kimi K3 | $3 in / $15 out per 1M | Cached input $0.30 |
| Dedicated H100 (80 GB) | $0.10833/min | About $6.50 per hour; T4 $0.01052/min, A100 $0.06667/min, B200 $0.16633/min |
What it does
- Model APIs
- Per-token endpoints for pre-optimised open models with cached-input pricing.
- Dedicated deployments
- Custom or open models on reserved GPUs, billed per minute from T4 to B200.
- Self-hosted and hybrid
- Run in Baseten Cloud, your own VPC, or both.
- Training
- Loops SDK for training, then deployment on the same infrastructure.
- Compliance
- SOC 2 Type II and HIPAA stated on the homepage.
Strengths and limitations
Strengths
- Clear path from shared Model APIs to dedicated GPUs for the same workload.
- Self-hosted and hybrid deployment options for data-control requirements.
- Published per-minute GPU prices.
- SOC 2 Type II and HIPAA compliance stated.
Limitations
- The per-token Model API catalogue is short compared with broad inference providers.
- Dedicated deployments bill per GPU-minute, which is costly for low-traffic hobby apps.
- Pro and Enterprise pricing is not published.
- The free credit amount for new accounts is not stated.
Who it suits
Good for
- Teams deploying fine-tuned or custom models to production
- Regulated companies that need HIPAA or VPC deployment
- Startups graduating from shared APIs to dedicated GPUs
Look elsewhere if
- Hobby projects with occasional traffic
- Developers who only need a wide menu of per-token models (a router is simpler)
Alternatives to Baseten
| Tool | From | Free option | Stages |
|---|---|---|---|
| Together AI Inference provider (open-weight models) | Usage-based | ||
| Claude API AI model API (model maker) | Usage-based | ||
| OpenAI API AI model API (model maker) | Usage-based | ||
| Amazon Bedrock Cloud AI model platform | — | ||
| Microsoft Foundry (Azure OpenAI) Cloud AI model platform | Usage-based | ||
| Cerebras Inference Inference provider (open-weight models) | Usage-based |
Questions
How much does an H100 cost on Baseten?
Baseten listed an 80 GB H100 at $0.10833 per minute (about $6.50 per hour) on 2026-09-29.
Can Baseten run in my own cloud?
Yes. Baseten offers self-hosted deployments in your own VPCs and hybrid setups alongside its managed cloud.
Is Baseten HIPAA compliant?
Baseten states SOC 2 Type II certification and HIPAA compliance on its homepage.
Sources read for this page