Together AI
Use a direct inference host while deciding when variable token usage justifies reserved capacity.
- Starting price
- Example USD 0.15 in / 0.60 out per 1M
- Free access
- No general free allowance verified
- Billing unit
- Model-specific tokens or media
- Platform
- Serverless APIs; dedicated options
- Verified
- 22 Oct 2026
Overview
What you provide
- Send a model-specific inference request or supply deployment requirements for reserved capacity.
What you get
- The service returns generated text or media with the selected model usage meter.
Choosing this tool
Together AI is useful when a production app needs a direct model host with both serverless and dedicated options. It also serves media models, but each family has a different meter. OpenRouter is a comparison for cross-provider routing, and DeepInfra for a second direct-host price and model-availability check.
A decision worked through
For the listed gpt-oss-120B example, 10 million input and 2 million output tokens cost USD 1.50 + USD 1.20 = USD 2.70 before tax. This is a token arithmetic example, not an evaluation of quality or a price for dedicated hardware.
Together AI alternatives for direct model inference and dedicated capacity
Compare Together AI with OpenRouter, DeepInfra for direct model inference and dedicated capacity. Match billing units and operating responsibility before comparing rates.
| Product | Price / unit | Free allowance | Input → output | Key limit |
|---|---|---|---|---|
| Together AIRead profileOfficial source ↗ | gpt-oss-120B USD 0.15/0.60 per 1M | Free allowance unverified | Model requests become text or media. | Dedicated services price separately. |
| OpenRouterRead profileOfficial source ↗ | Standard: model cost + 5.5% | 50 free-plan requests/day | Messages become routed responses. | Provider policies still differ. |
| DeepInfraRead profileOfficial source ↗ | Llama 3.3 70B USD 0.10/0.32 per 1M | Free quota unverified | Prompts become metered model output. | Priority 1.5x; Flex 0.8x. |
Setup & workflow
- Pin a supported model and confirm both input and output rates.
- Run an application evaluation with recorded token counts.
- Compare sustained usage with dedicated options only after collecting real demand.
Requirements & installation
- Configure a Together API key and choose a model currently listed for the needed modality.
Limits & review
- Price and availability are model-specific, and dedicated services can have commitment or uptime costs. A media default-size estimate is not a universal per-output quote.
Your part
- Measure answer quality, latency and errors with representative application inputs.
Sources & verification4
This profile is based on official sources, not a hands-on product test.
Vendor descriptions and demos document advertised features. Editorial guidance is based on these sources.
- Official Together AI functionalitywww.together.ai
- Official Together AI pricingwww.together.ai
- OpenRouter official referenceopenrouter.ai
- DeepInfra official referencedeepinfra.com
