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
There's a free plan; paid tiers are quoted.
Together AI lists a free plan, but does not publish its allowance or limits. For a team of one to five, there is no fixed monthly or annual subscription to budget for: Serverless Inference charges by tokens, while Dedicated Inference charges by GPU-hour. The same usage rates apply whether one person or five people are making requests. Before estimating a bill, choose a model and forecast input and output tokens; add GPU or media services only if the workload needs them.
01Every plan, line by line
| Plan | Billed annually | Billed monthly | Built for |
|---|---|---|---|
| Serverless Inference | — | Quote | Teams serving models on demand |
| Batch Inference | — | Quote | Teams processing batch workloads |
| Provisioned Throughput | — | Quote | Production teams reserving capacity |
| Dedicated Inference | — | Quote | Teams needing dedicated GPU infrastructure |
| Dedicated Container Inference | — | Quote | Teams deploying generative media models |
Add-ons and overages
- Image generation examples: $0.0154 per image for FLUX.2 dev; $0.053 per image for GPT Image 2.
- Video generation examples: $0.115 per video for ByteDance Seedance 2.5; $0.57 per video for ByteDance Seedance 1.0 Pro.
- Audio examples: $0.0015 per audio minute for Whisper Large v3; $65 per 1M characters for Cartesia Sonic-3.
- Displayed image prices use listed default steps; additional costs apply when exceeding default steps. Actual prices may vary by settings.
02What the plan grid doesn't tell you
Token usage adds up
Serverless and Batch Inference charge by model and token type. Estimate input and output separately; the listed rates vary by model.
GPU-hour charges
Dedicated Inference lists H100 at $5.49 and B200 at $8.99 per GPU-hour. Other listed GPU types and reserved rates require contacting sales.
Media generation
Image, video and audio generation have separate rates. Image prices use listed default steps; additional costs apply when exceeding them.
Free-plan limits
The free plan is listed, but its allowance and eligible models are not published. Ask Together AI what is included and what happens at the limit.
03Which plan should you buy?
| If you are… | Get | Why |
|---|---|---|
| Solo builder | Serverless Inference | Start with on-demand model use and pay per token; check the free-plan limits before relying on them. |
| Team of 3 | Serverless Inference | There is no listed seat fee, so budget by model and token volume rather than multiplying a per-user price. |
| Team of 5-10 | Batch Inference | Consider it for batch workloads; the listed offering supports up to 30 billion tokens per model. |
04Is Together AI worth the money?
Together AI can suit a small team that wants to try open-source models without buying GPU infrastructure or signing up for a seat-based subscription. The listed free plan may be enough for early evaluation, but the vendor does not publish its limits, so confirm what it covers first. Paid use is priced by tokens or GPU time, not by team size. That flexibility helps when demand varies, but it makes forecasting and monitoring usage part of the budget work.
Together AI pricing FAQ
Does Together AI have a free plan?
Yes. Together AI lists a free plan, but its allowance, eligible models and other limits are not published. Ask the vendor for the current terms before using it to estimate a launch budget.
How much does Together AI cost per month?
The pricing page lists usage-based token and GPU-hour prices, but no monthly or annual subscription price. A monthly bill depends on model selection and usage, so Together AI does not publish one standard monthly total.
Does Together AI charge per seat?
No per-seat price is listed. The listed charges are based on usage, such as tokens for inference or GPU-hours for Dedicated Inference. A team of three or five therefore cannot calculate its bill by multiplying a seat rate.
What is the cheapest paid Together AI option?
Serverless Inference is the direct usage-based option for teams serving models on demand. Its price varies by model and input or output tokens. The pricing page lists some models with zero-dollar rates, but a team should confirm availability and terms before treating them as a paid-plan estimate.
