Serverless inference
Route requests to the fastest healthy endpoint with automatic batching, speculative decoding, and per-token load balancing.
AI Infrastructure Cloud
The cloud built for frontier AI.
Together deploys, scales, and serves frontier models on dedicated GPU clusters and a global inference edge — one manifest, minutes to production.
2xFaster inference vs. baseline
60%Lower GPU cost at scale
90%Less time on infra ops
Platform
Everything you need to serve models at scale
A full control plane for inference — from the first request to planet-wide throughput.
Elastic GPU clusters
Reserve H-class accelerators in seconds and scale from one node to thousands with live migration and zero-downtime rollovers.
Global serving edge
Serve from 30+ regions with latency-based routing, so every user sits within 20ms of a warm model replica.
How it works
From repo to global endpoints in minutes
Connect
Link your model weights, container, or Hugging Face repo. Weights sync automatically and cache in every region.
Configure
Declare replicas, autoscaling rules, and GPU type in one versioned, reviewable YAML manifest.
Deploy
Ship to global endpoints in minutes with canary rollouts, instant rollback, and live traffic dashboards.
Customers
Teams shipping on Together
We cut inference spend by 58% in the first month and nobody touched an ops runbook. It just scales.
VP Engineering, VectorLabs
Rollouts that took our old platform an afternoon now finish before standup. The edge routing is unfair.
CTO, Northwind AI
Pricing
Simple pricing for any scale
Start free, scale predictably, and go custom when compliance demands it.
$0/month
- 3M tokens / month included
- Community model library
- Shared GPU capacity
- API keys & playground
$249/month
- Dedicated GPU pools
- Autoscaling to 100+ replicas
- Global edge, 30+ regions
- 99.99% uptime SLA
CustomVolume pricing
- BYO cloud & VPC peering
- Custom model fine-tuning
- SOC 2 & HIPAA compliance
- 24/7 dedicated engineer
FAQ
Frequently asked questions
Which models can I deploy on Together?
Any weights you own or license: open-weight models like Llama, Mistral, and Qwen, plus your own fine-tunes. Bring safetensors, a container, or a Hugging Face repo — we handle the rest.
How does billing work?
You pay a monthly rate for reserved capacity and per-request pricing for serverless inference. Annual plans save 20%, and unused serverless credits roll over for 90 days.
Is my data private?
Yes. Prompts and completions are never used for training, are encrypted in transit and at rest, and can be pinned to single-region processing for compliance workloads.
Get Started
Talk to sales
Ship frontier AI, together.
Join thousands of teams serving production traffic on Together. Request access and we'll provision your first cluster today.