We may earn a commission if you buy through links on this page. It never changes a score or a ranking. How we make money
Pricing is quote-based.
Together AI has no published monthly starting price: Serverless Inference and Batch Inference are usage-based, while Dedicated Inference is billed per GPU-hour. The pricing page lists Llama 3 8B Instruct Lite at $0.14 for input and output per 1M tokens, and H100 dedicated inference at $5.49 per GPU-hour on demand. Provisioned Throughput has an example estimate of $21,600 monthly for 10 PTUs. There is no free plan listed, and usage totals depend on workload.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Serverless Inference | — | Quote | Teams using hosted model APIs |
| Batch Inference | — | Quote | Teams running batch workloads |
| Provisioned Throughput | — | Quote | Teams reserving dedicated token capacity |
| Dedicated Inference | — | Quote | Teams needing single-tenant GPU inference |
Add-ons and overages
- Image pricing includes default steps; extra costs apply beyond those steps, but amounts are not listed.
02What the plan grid doesn't tell you
Image step charges
Image pricing includes default steps. Extra steps cost more, but the pricing page does not list the amounts.
PTU capacity varies
Capacity per PTU depends on model and token type, so the $21,600 monthly estimate for 10 PTUs may not fit every workload.
Batch fees are usage-based
Batch API rates depend on model usage. No separate batch discount or plan fee is published.
Reserved GPU rates
Reserved pricing is not listed; the pricing page directs buyers to contact sales.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Teams calling hosted models | Serverless Inference | Pay by usage rather than a listed monthly plan fee; model rates vary, and some list $0.00 input and output. |
| Teams reserving token capacity | Provisioned Throughput | The pricing page gives a 10-PTU estimate of $21,600 monthly; confirm capacity for your model and token type. |
| Teams needing single-tenant GPUs | Dedicated Inference | On-demand H100 is listed at $5.49 per GPU-hour; reserved pricing and several GPU models require a sales quote. |
04Is Together AI worth the money?
Together AI offers several ways to run open models, but the pricing is not a simple subscription you can budget from a monthly plan grid. Serverless and batch costs depend on usage; reserved token capacity has a substantial example estimate, and dedicated GPU rates are hourly or quote-based. It may suit teams that can forecast workloads and choose the right deployment model. We would estimate token volume, capacity and GPU hours before committing; the listed rates alone do not predict a total bill.
Together AI pricing FAQ
Is Together AI free to use?
The pricing information does not list a free plan, and a free trial is not published. Some models list $0.00 input and output, but that does not establish that the platform as a whole is free. Check the rate for the specific model and service you plan to use.
How much is Together AI valued at?
A company valuation is not published in the published pricing information. Together AI’s listed usage rates and GPU-hour prices are service costs, not a valuation. We would need a dated company announcement or other reliable source to give a valuation figure.
How much do AI subscriptions cost?
There is no universal subscription price in the Together AI pricing information: its services are usage-based rather than listed monthly subscriptions. Examples include $0.14 for input and output per 1M tokens for Llama 3 8B Instruct Lite and $5.49 per H100 GPU-hour on demand.
Is AI getting more expensive?
The pricing information here does not show whether AI prices are rising or falling over time. Together AI lists usage-based model rates, hourly GPU prices and a monthly PTU estimate, but those figures do not establish a broader price trend. Compare current rates for your workload.