Together AI
4.4
Together AI provides serverless inference, dedicated endpoints, GPU clusters, fine-tuning and a code sandbox for open-source models. Serverless token rates range widely by model; dedicated H100 is $5.49 a GPU-hour on demand, and GPU clusters are $3.99 for H100 on demand or $1.99 preemptible, with reserved H100 from $3.19 depending on commitment.
- Commercial use OK
- Enterprise contract
- Pricing
- 従量課金
- Free tier
- No
- Commercial use
- Permitted
- Japanese UI
- Not confirmed
- Enterprise contract
- Available
What it is good for
- Running open-source models without operating GPUs
- Workloads that need dedicated throughput rather than shared
- Fine-tuning open models at published per-token rates
- Comparing several open models through one API
Strengths and trade-offs
Strengths
- Published GPU-hour rates including preemptible and reserved tiers
- Covers serverless through to dedicated clusters as usage grows
- Fine-tuning priced per training token with a stated minimum
Trade-offs
- No clearly described ongoing free tier
- Pricing complexity across serverless, dedicated and clusters
- Open models still trail frontier closed models on hard tasks
Editors' review
Transparent enough to actually plan with
The preemptible H100 rate at $1.99 is the interesting number. For batch work that can tolerate interruption, it halves the cost of the single largest line item in most inference budgets.
The breadth of options is a strength once you know your workload shape and a source of confusion before that.
Verdict
A good home for open-model workloads at scale. Start serverless, move to clusters when the maths flips.
Compiled from official documentation and public sources, with our own editorial judgement. We do not publish figures we cannot source.